Olivier de La Grandville-Bond Pricing and Portfolio Analysis-The MIT Press (2000)

M
d
.

d
a
l
i
m

#
5
3
3
8
4
7

8
/
9
/
0
0

C
y
a
n

M
a
g

Y
e
l
o

B
l
a
c
k
BOND PRICING AND PORTFOLIO
ANALYSIS
BOND PRICING AND PORTFOLIO
ANALYSIS
Protecting Investors in the Long Run
Olivier de La Grandville
The MIT Press
Cambridge, Massachusetts
London, England
: 2001 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any electronic or
mechanical means (including photocopying, recording, or information storage and retrieval)
without permission in writing from the publisher.
This book was set in Times New Roman on `3B2' by Asco Typesetters, Hong Kong.
Printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
La Grandville, Olivier de.
Bond pricing and portfolio analysis : protecting investors in the long run /
by Olivier de La Grandville.
p. cm.
Includes bibliographical references and index.
ISBN 0-262-04185-5
1. BondsPrices. 2. Interest rates. 3. Investment analysis. 4. Portfolio
management. I. Title.
HG4651.L325 2000
332.63
/
23dc21 00-058682
For my wife, Ann, and our children,
Diane, Isabelle, and Henry,
and in loving memory of my mother,
Anika Pataki
CONTENTS
Introduction ix
1 A First Visit to Interest Rates and Bonds 1
2 An Arbitrage-Enforced Valuation of Bonds 25
3 The Various Concepts of Rates of Return on Bonds: Yield to
Maturity and Horizon Rate of Return 59
4 Duration: Denition, Main Properties, and Uses 71
5 Duration at Work: The Relative Bias in the T-Bond Futures
Conversion Factor 95
6 Immunization: A First Approach 143
7 Convexity: Denition, Main Properties, and Uses 153
8 The Importance of Convexity in Bond Management 171
9 The Yield Curve and the Term Structure of Interest Rates 183
10 Immunizing Bond Portfolios Against Parallel Moves of the
Spot Rate Structure 209
11 Continuous Spot and Forward Rates of Return, with Two
Important Applications 221
12 Two Important Applications 237
13 Estimating the Long-Term Expected Rate of Return, Its
Variance, and Its Probability Distribution 267
14 Introducing the Concept of Directional Duration 295
15 A General Immunization Theorem, and Applications 307
16 Arbitrage Pricing in Discrete and Continuous Time 327
17 The Heath-Jarrow-Morton Model of Forward Interest
Rates, Bond Prices, and Derivatives 359
18 The Heath-Jarrow-Morton Model at Work: Applications to
Bond Immunization 383
By Way of Conclusion: Some Further Steps 405
Answers to Questions 411
Further Reading 437
References 441
Index 447
viii Contents
INTRODUCTION
Duke of Suffolk. I will reward thee for this venturous deed.
Henry VI
The last two decades have witnessed an unusually high volatility of
interest rates, making the choice of so-called xed-income instruments a
dicult but potentially rewarding task. The reasons for such increasing
volatility can be traced back to the end of the system of xed exchange
rates set up at Bretton Woods. Perhaps the dividing date in the postwar
period is August 15, 1971, when the United States announced that it
would not commit itself further to selling gold to foreign central banks at
$35 per ounce. Since that date, foreign exchange markets have been
shaken by high turbulence, which has had direct consequences on the price
of loanable funds throughout the world.
It may be useful to remember how the whole problem started. In the
twenty years following the end of World War II, the U.S. dollarafter
being scarce for some timebecame relatively abundant because of the
increasing decit of the American balance of payments. This decit had
three main causes: relatively higher ination in the United States led to a
decit of the U.S. trade balance; direct investment made by American
rms abroad far outweighed foreign direct investments in the United
States; and important aid abroad contributed to increasing the decit. In
order to maintain the quasi-xed parities of their currencies with the U.S.
dollar, foreign monetary authorities were compelled to intervene in the
foreign exchange markets in order to buy the excess supply of dollars; this
excess supply was quickly compounded by expectations of a downfall
of the system when it appeared, in the early 1970s, that the amount of
foreign exchange held by foreign central banks was nearly four times the
amount of the gold reserves held by the United States.
In the years that followed, the major economic powers were unable to
set up a durable system of xed exchange rates. As a result, huge capital
movements, anticipating increases or falls in exchange rates, led to strong
uctuations in rates of interest. And monetary authorities, in order to pro-
tect the external value of their currencies or to prevent them from rising,
took a number of measures that aected those interest rates. Such volatility
was increased by two major factors: the two oil shocks of 1973 and 1979,
and ination spells, varying in intensity from country to country. Finally,
we should keep in mind that today most countries, following the example
set by the United States in 1979, have decided to dene their monetary
policies on the basis of specied monetary targets in lieu of pursuing the
even more illusory objective of xing interest ratesa denite blow to
stability.
Analytical background
In nance, as in any science, entirely novel ideas have been relatively
scarce, but each time they appeared they have had a profound inuence
on our way of thinking, and wide applications. In this study on the
valuation of bonds, we will make use of some remarkable discoveries that
span more than two centuries: they range from Turgot de L'Aulne's Re-
exions (1766) to the Heath-Jarrow-Morton (1992) stochastic model of
the interest rate term structure. Let us now say a brief word about those
developments.
To the best of our knowledge, we owe to Turgot de L'Aulne the rst
explanation of the equilibrium price dierence between a high-quality as-
set with given risk and a low-quality asset bearing the same risk. Speci-
cally, his purpose was to determine the equilibrium prices of good lands
and bad lands. In his Reexions sur la formation et la distribution des
richesses1 he gave the correct answer: through arbitrage, from an initial
disequilibrium situation that would enable arbitrageurs to step in, prices
would move to levels such that the expected return on both assets (on both
types of land) would be the same. He added that the yieldsand conse-
quently the priceson all types of investments bearing dierent risks
were linked together (in his days, the asset bearing the least risk, land
ownership, was followed by loans, and nally by assets in trade and in-
dustry). The equilibrium levels of the expected returns were separated by
a risk premium, as the levels of two liquids of dierent densities in the two
branches of a siphon would always remain separated from each other; one
1A. Turgot de L'Aulne, Reexions sur la formation et la distribution des richesses, 1776.
Reprinted by Dalloz, Paris, 1957. See particularly propositions LXXXII through XCIX,
pp. 128139.
x Introduction
level could not increase, though, without the other one moving up as
well.2
It would take more than a century for Turgot de L'Aulne's basic idea,
corresponding to a given risk level, to be formalized into an equation,
which we owe to Irving Fisher (1892).3 Fisher showed that in a risk-free
world, equilibrium prices corresponded to the equality between the rate of
interest and the sum of the direct yield of the capital good and its relative
increase in price. Fisher had obtained his equation through an `ìntegral''
approach, by equating the price of an asset to the sum (an integral in
continuous time) of the discounted cash ows of the asset and dierentiat-
ing the equality with respect to time. It is interesting to note that, to the
best of our knowledge, the rst time this equation was obtained through a
dierential, and an arbitrage, point of view was in Robert Solow's fun-
damental 1956 ``Contribution to the theory of economic growth.''4
The power and generality of Fisher's equation can hardly be overstated.
Since optimal economic growth theory is the theory of optimal investment
over time, it is of no surprise that the Fisher equation is central to the
solution of variational problems in growth theory. For instance, a typical
problem of the calculus of variations is to determine the optimal path of
investment to maximize a sum, over time, of utility ows generated in a
given economy; the problem can be extended to the optimal allocation of
exhaustible resources. It can be shown that the Euler equations governing
the optimal paths of renewable stocks of capital or those of nonrenewable
resources are nothing else than extensions of the Fisher equation.5
2We cannot resist the pleasure of quoting Turgot in his beautiful eighteenth-century style:
``Les dierents emplois des capitaux rapportent donc des produits tres inegaux : mais cette
inegalite n'empeche pas qu'ils s'inuencent reciproquement les uns les autres et qu'il ne
s'etablisse entre eux une espece d'equilibre, comme entre deux liquides inegalement pesants,
et qui communiqueraient ensemble par le bas d'un siphon renverse, dont elles occuperaient
les deux branches ; elles ne seraient pas de niveau, mais la hauteur de l'une ne pourrait
augmenter sans que l'autre montat aussi dans la branche opposee'' (Turgot de L'Aulne,
1776, op. cit., p. 130).
3Irving Fisher's original work on the equation of interest was published in Mathematical
Investigations in the Theory of Value and Price (1892); Appreciation and Interest (1896); in
``Reprints of Economic Classics'' (New York: A. M. Kelley, 1965).
4Robert M. Solow, `À contribution to the theory of economic growth,'' The Quarterly
Journal of Economics, 70 (1956), pp. 439469.
5The interested reader may refer to Olivier de La Grandville, Theorie de la croissance econ-
omique (Paris: Masson, 1977), and to ``Capital theory, optimal growth and eciency condi-
tions with exhaustible resources,'' Econometrica, 48, no. 7 (November 1980), pp. 17631776.
xi Introduction
In the present study, the Fisher equation will quickly be put to work.
We will ask the following question: suppose that interest rates do not
change within a year (or, equivalently, that within a year, interest rates
have come back to their initial level) and that interest rates are described
by a at structure. Consider a coupon-bearing bond without default risk,
with a remaining maturity of a few years; suppose, nally, that the bond
was initially, and therefore still is after one year, under par. At what
speed, within one year, has it come back toward par (in other words, what
was the relative change in price of the bond over the course of one year)?
We will show that the Fisher equation gives an immediate answer: the
relative rate of change will always be equal to the dierence between the
rate of interest and the bond's simple (or direct) yield, dened as the ratio
between the coupon and the bond's initial value.
A central question, however, set up by Turgot remained yet unsolved:
for any two assets bearing dierent risks, what are the equilibrium risk
premia born by each asset which lead to equilibrium asset prices? A rst
answer to this question would have to wait for nearly two centuries: it was
not until the 1960s that William Sharpe and Jack Lintner would develop
independently their famous capital asset pricing model on the basis of
optimal portfolio diversication principles set by Harry Markovitz about
a decade earlier. The importance of their discovery is that it yields, at the
same time, a measure of risk and the asset's risk premium.
It would take the development of organized markets of derivative
products for other major advances to be made: there was rst the Black-
Scholes6 and Merton7 valuation formula of European options (1973);
then the recognition by Harrison and Pliska8 (1981) that the absence of
arbitrage was intimately linked to the existence of a martingale probabil-
ity measure. And a major discovery was nally made by David Heath,
Robert Jarrow, and Andrew Morton in 1992;9 it is of central importance
to our subject, since it deals with the stochastic properties of the term
6Fischer Black and Myron Scholes, ``The pricing of options and corporate liabilities,''
Journal of Political Economy, 81 (1973), pp. 637654.
7Robert Merton, ``Theory of rational option pricing,'' Bell Journal of Economics and Man-
agement Science, 4 (1973), pp. 141183.
8Michael Harrison and Stanley Pliska, ``Martingales and stochastic integrals in the theory of
continuous trading,'' Stochastic Processes and their Applications, 11 (1981), pp. 215260.
9David Heath, Robert Jarrow, and Andrew Morton, ``Bond pricing and the term structure
of interest rates: a new methodology for contingent claims valuation,'' Econometrica, 60
(1992), pp. 77105.
xii Introduction
structure of interest rates. Heath, Jarrow, and Morton's fundamental
discovery is the following: arbitrage-free markets imply that, if a Wiener
process drives the forward interest rate, the drift term of the stochastic
dierential equation cannot be independent from its volatility term; on
the contrary, it will be a deterministic function of the volatility. Today
this surprising result is central to bond pricing and hedging in the context
of a stochastic term structure.
An overview of the book
Volatility of interest rates has direct consequences on the returns of bonds,
because of both a price eect and an eect on the reinvestment of cou-
pons. This naturally leads the money manager to play two dierent roles.
Often he will choose to have an active style of management, trying to use
maximum information to forecast interest rates and take positions on the
bond market; or he can try to protect the investor from the up-and-down
swings of rates of interest and try to immunize bond portfolios against
those changes.
Our purpose in this book is to discuss some of the most important
concepts pertaining to bond analysis and management, with major em-
phasis on the protection of investors against changes in interest rates. In
our introductory chapter, we pay a rst visit to bonds and interest rates.
In particular, we explain how the various types of bonds are quoted in the
nancial press by dening basic concepts related to those xed-income
securities, and we introduce the various ways of dening interest rates.
But we also ask the rather innocuous question: why do interest rates exist
in the rst place? The concept of forward rates gives us an opportunity to
meet the concept of arbitrage. In the nal part of our rst chapter, we
describe the main features of a particularly important nonstandard type
of bondthe convertible. In chapter 2, we take up the valuation of a bond.
One special feature of this chapter is to relate the valuation of bond to the
arbitrage equation of interest; in particular, we show that the price of the
bond and its rate of change must be such that this equation always holds.
In chapter 3, we survey the most relevant measures of rates of return on
bonds, namely the yield to maturity and the horizon rate of return.
We are then able to turn to the central concept of duration, its main
properties and uses (chapters 4 and 5). The all-important concept of
xiii Introduction
immunization (protection of the investor against uctuations in interest
rates) is introduced in chapter 6. In chapters 7 and 8, the concept of con-
vexity is introduced and related to that of duration. With concrete exam-
ples, we show its role in bond management. Chapter 9 deals with the
question of the term structure of interest rates and its signicance in esti-
mating what the future structure may look like.
The fundamental result of immunization of bond portfolios for parallel
changes in the term structure of interest rates is demonstrated in chapter
10. In the next chapter, we discuss spot and forward rates of return in
continuous time and apply the central limit theorem to justify the work-
horse of today's nancial theory, the lognormal distribution of the dollars
returns and asset prices. This leads us, in chapter 12, to a theoretical jus-
tication of the lognormal distribution of asset prices and to an exami-
nation of recent methods used to determine the term structure of interest
rates. In chapter 13, we take up a subject that seems to us little known and
particularly important to bond investors: estimating the long-term
expected rate of return, its variance, and its probability distribution. We
introduce in chapter 14 a new concept, directional duration, which gen-
eralizes the Macaulay and Fisher-Weil concepts.
We then oer, in chapter 15, a general immunization theorem, which
we put to the test of any change received by the term structure of interest
rates. This theorem generalizes the result demonstrated in chapter 10.
Arbitrage in discrete and continuous time, which is so central to modern
nancial analysis, is described in chapter 16. Here we present the concept
of probability measure change. In chapter 17, we discuss the famous
Heath-Jarrow-Morton model of forward interest rates and bond prices,
and in chapter 18 we put this model to work by applying it in some detail
to bond portfolio immunization.
The target audience
We have tried to keep the level of technicality of this study as low as
possible. For the most part, mathematical developments have been given
in appendixes, and only quite simple ones remain in the main text. The
rst ten chapters can be used for core studies in an undergraduate course;
the last part of the book, which requires some familiarity with probability
and stochastic calculus, can be used together with a quick review of the
rst part for a graduate course.
xiv Introduction
What is new in this book?
We can promise a systematic valuation of bonds through arbitrage as well
as some (nice) surprises and hopefully novel results.
.
For instance, the value of a bond can always be expressed as a weighted
average between 1 and the ratio of the coupon rate to the rate of interest,
with weights that are moving through time. The formula we suggest may
be very useful in understanding how the bond's value will evolve over
time toward par for any given rate of interest (chapter 2).
.
An explanation for the apparently strange phenomenon of the indeter-
minacy between an increase in a bond's maturity and its duration; we will
also give an economic interpretation of the value of the duration of a
perpetual bond, whatever its coupon rate may be (chapter 4).
.
An introduction of the concept of the relative bias regarding the most
important contract an nancial futures markets, the Chicago Board of
Trade T-bond futures contract (chapter 5).
.
The importance of convexity in bond management and ways of
improving the convexity of a bond portfolio (chapters 7 and 8).
.
A proof of the immunization theorem for parallel changes in the spot
rate curve, based on a variational approach (chapter 10).
.
A discussion of the exact relationship between spot and forward rates of
return. In particular, we show that a spot rate structure cannot increase
and reach a plateau without the forward rate decreasing on an interval
before that stage (chapter 11).
.
Recent methods in deriving the structure of interest rates from coupon
bond prices (chapter 12).
.
Exact formulas for the expected value and variance of the long-term
rate of return on bonds and, more generally, on any investment, directly
expressed as functions of one-year estimates, and an explanation of the
probability distribution governing the one-year return and the long-
horizon return (chapter 13).
.
A new concept of duration, based on the directional derivative: direc-
tional duration (chapter 14). We use this to cope with nonparallel shifts in
the spot rate structure.
xv Introduction
.
A general immunization theorem, applicable to any kind of shock the
spot rate curve may receive (chapter 15).
.
A geometrical interpretation of a martingale probability construction
(chapter 16).
.
An intuitive demonstration of the Cameron-Martin-Girsanov theorem
(chapter 16).
.
A complete discussion of the one-factor Heath-Jarrow-Morton model
of interest term structure and its applications to bond portfolio duration
and immunization. In particular, we give the full demonstration of the
HJM one-factor model in the all-important case of the Vasicek volatility
coecient (chapter 17).
.
We discuss in detail how the HJM model duration and immunization
results dier fromand generalizethe classical results (chapter 18).
In order for this text to be as user-friendly as possible, each chapter is
followed by a series of questions, problem sets, and projects, and detailed
solutions for all of these can be found at the back of the book.
Acknowledgments
Many individuals, by their comments and support, have been very helpful
in the completion of this book. It is a pleasure for me to thank the part-
ners at Lombard, Odier & Cie in Geneva, in particular Jean Bonna,
Thierry Lombard, Patrick Odier, and Philippe Sarrasin, their xed in-
come group in Geneva, Zurich, and London, and especially Ileana Regly,
for their support. I also thank my colleagues Christopher Booker, Samuel
Chiu, Ken Clements, Jean-Marie Grether, Roger Guerra, Yuko Hara-
yama, Roger Ibbotson, Blake Johnson, Paul Kaplan, Rainer Klump,
David Luenberger, Michael McAleer, Paul Miller, Anthony Pakes,
Elisabeth Pate-Cornell, Jacques Peyriere, Elvezio Ronchetti, Wolfgang
Stummer, James Sweeney, Meiring de Villiers, Nicolas Wallart, Juerg
Weber, and Pierre-Olivier Weill. Special thanks are due to Eric Knyt, who
has read the entire manuscript with his usual eagle eye and has made
many useful suggestions. And I would like to extend my deepest appreci-
ation to Mrs. Huong Nguyen, for her admirable typing, cheerfulness, and
her devotion to her job.
xvi Introduction
Parts of this book were written while I was visiting the Department
of Management Science and Engineering (formerly the Department of
EngineeringEconomics and Operations Research) at Stanford Univer-
sity, and the Department of Economics at the University of Western
Australia. I would like to extend my appreciation to both institutions for
their wonderful working environment and friendly atmosphere.
Finally, it is a pleasure to express my thanks to the sta at MIT Press
and to Peggy M. Gordon for their eciency and genuine care in produc-
ing this book.
About the author
Olivier de La Grandville is Professor of Economics at the University of
Geneva. Since 1988 he has been a Visiting Professor at the Department
of Management Science and Engineering at Stanford University. Profes-
sor de La Grandville is the rst recipient of the Chair of Swiss Studies at
Stanford; his teaching and research interests range from microeconomics
to economic growth and nance. The author of six books in these areas,
his papers have been published in journals such as The American Eco-
nomic Review, Econometrica, and the Financial Analysts Journal. He has
also held visiting positions at the following institutions: Massachusetts
Institute of Technology, University of Western Australia, University of
Lausanne, and Ecole Polytechnique Federale de Lausanne. He is a regu-
lar lecturer to practitioners.
Although, for no apparent reason, he has not yet made the team, his
favorite baseball club is still the San Francisco Giants.
xvii Introduction
1
A FIRST VISIT TO INTEREST RATES
AND BONDS
Shylock. . . . and he rails . . . on my well-won thrift
Which he calls interest.
The Merchant of Venice
Macbeth. Cancel and tear to pieces that great bond
Which keeps me pale.
Macbeth
Chapter outline
1.1 A rst look at interest rates or rates of return 3
1.1.1 Horizons shorter than or equal to one year 4
1.1.2 Horizons longer than one year 5
1.1.2.1 Horizons longer than one year and integer numbers 5
1.1.2.2 Horizons longer than one year and fractional 5
1.2 Forward rates and an introduction to arbitrage 6
1.3 Creating synthetically a forward contract 9
1.4 But why do interest rates exist in the rst place? 10
1.5 Kinds of bonds, their quotations, and a rst approach to yields on
bonds 13
1.5.1 Treasury bills 13
1.5.2 Treasury notes 17
1.5.3 Treasury bonds 18
1.6 Bonds issued by rms 18
1.7 Convertible bonds 19
Overview
A bond is a certicate issued by a debtor promising to repay borrowed
money to a lender or, more generally, to the owner of the certicate, at
xed times. The following glossary will be immediately useful.
.
The amount borrowed is called the principal; it is often referred to as the
par value, or face value, of the security.
.
The time span until reimbursement will be called the maturity of the
bond. Normally, ``maturity'' designates the date at which the principal
is due; in this book, however, we will often consider maturity to be the
dierence between the date of reimbursement and today.
.
The contract underlying the security may involve a number of pay-
ments of equal size. Those cash ows to be received at regular intervals
by the lender are called coupons; they may be paid yearly (as is the case
frequently in Europe), twice a year, or on a quarterly basis (as in the
United States).
.
The coupon rate is the yearly value of cash ows paid by the bond issuer
divided by the par value.
.
If no coupons are to be paid, the bond is called a zero-coupon bond. A
zero-coupon bond is said to be a pure-discount security. Notice that all
coupon-bearing bonds automatically become pure-discount securities
once the last coupon and par value are all that remain to be paid out by
the issuer.
If the borrower does not default on his promises, the lender can count on
xed payments throughout the lifetime of the security. But even in this
case, owning such a nancial instrument does not guarantee a xed yearly
yield on invested money.
The rst and foremost reason yields on bonds are likely to uctuate
comes from an essential property of the bond: as a promise on future cash
ows, it has a value that is most likely to move up or down, because each
of those promises can uctuate in valuea fact which, in turn, needs to
be explained.
Suppose a large company with an excellent credit rating (soon to be
dened) borrows today at a 5% coupon rate by issuing 20-year maturity
bonds. The company has done this in a nancial market where a supply of
loanable funds meets a demand for such funds; presumably, the numbers
of lenders and borrowers are such that none of them, individually, is able
to have a signicant impact on this market. We can reasonably imagine
that if the company has tried to obtain, and has been able to secure, a 5%
loan in that peculiar securities market, it is because:
1. It would have been unable to obtain it at a lower coupon rate (4.75%
for instance); competition from other borrowers for relatively scarce
loanable funds has pushed up the coupon rate to 5%.
2. It would have been foolish for rms to oer a higher coupon rate; for a
5.25% coupon rate, borrowers would have faced an excess supply of
loanable funds, which would have driven down the coupon rate to 5%.
Suppose now, a few weeks later, conditions on that nancial market
have considerably changed. Due to good economic prospects, demand for
loans from the same category of rms has increased; however, supply has
2 Chapter 1
not changed. On the 20-year maturity market, the coupon rate is bound
to increase, perhaps to 5.50%. Obviously, something momentous will
happen to our 5% coupon rate security; its price will fall. Indeed, there
exists a market for that bond, and the initial buyer may want to sell it. It
is evident, however, nobody will want to buy a ``5% coupon'' at par when
for the same price the buyer can get a ``5.5%.'' Therefore, both the seller
and the buyer of the previously issued security may agree on a price that
will be lower than par. If a few days later market conditions change
again (this time in the opposite direction, perhaps because large inows of
capital movements from abroad have increased the supply of loanable
funds on that market, entailing a decrease in the coupon rate from 5.5%
to 5.25%), the price of the security is bound to move again, this time
upward.
We will determine in chapter 2, on bond valuation, exactly how large
those uctuations will be. For the time being, our aim is to recognize that
bond values will move up or down according to conditions of the nancial
market, and this creates the rst factor of uncertainty for the bond holder.
To understand a second reason for uncertainty, suppose that our
investment horizon, in buying the 20-year bond, is a few yearsor even
20 years. The various cash ows received will be reinvested at rates we do
not know today. So even if we buy a bond at a given price and hold onto
it until maturity, we cannot be certain of the yield of that investment. This
uncertainty is referred to as reinvestment risk.
There are thus two reasons why assets that do not carry any default risk
are still prone to uncertainty. First, even for zero-coupon bonds there is
some price risk if bonds are not held until maturity; second, there is re-
investment risk. These two sources of risk nd their origins in the general
conditions of supply and demand of loanable funds, which usually are
liable to change, in most cases in an unpredictable way. Naturally, some
fund managers will try to know more, and quicker, than other managers
about such changes in order to benet from bond price increases, for
instance.
1.1 A rst look at interest rates or rates of return
Until now we have not yet referred to, much less dened, rate of interest.
The dierent concepts of rate of interest can be distinguished according to
3 A First Visit to Interest Rates and Bonds
.
the time of the inception of the contract;
.
the time at which the loan begins;
.
the time at which the loan ends;
.
the mode of calculation of the interest between the beginning and the
end of the loan.
A loan agreed upon on a certain date, starting at that date will give rise
to a spot interest rate. A loan starting after the date of the contract is a
forward contract. In the following discussion we will refer to the horizon
as the length of time between the start of the loan and the time of its
reimbursement. This horizon will also be referred to as the trading period.
The easiest way to dene a spot rate is to say it is equal to the (yearly)
return of a zero-coupon with a remaining maturity equal to the investor's
horizon. Therefore, it is the yearly rate of return that transforms a zero-
coupon value into its par value. Unfortunately, this denition is not quite
precise, because there are many dierent ways to calculate a ``per year
rate.'' Generally, however, the following rules are followed: Call i the rate
of interest, B
t
the value of the zero-coupon bond at time t, and B
T
the par
value of the zero-coupon bond (or the reimbursement value of the bond at
time T). The investor's horizon is h = T t.
1.1.1 Horizons shorter than or equal to one year
For horizons T t equal to or shorter than one year, the rate of interest is
the relative rate of increase between the bond's value at t, B
t
, and the par
value B
T
, divided by the horizon's length T t. This is the simplest de-
nition that can exist, since it corresponds to capital gain per time unit (per
year, if the time unit is a year). For example, suppose a zero-coupon with
par value equal to B
T
= 100 is worth 98 at time t (today, three months
before maturity). The rate of interest for the three-month period,
expressed on a yearly basis, is
i =
B
T
B
t
B
t
_
(T t) =
100 98
98
_
1
4
= 8X163% per year (1)
Notice there is an alternate way to dene the same concept: rearrange the
above expression and write it as
i(T t) =
B
T
B
t
B
t
4 Chapter 1
or, equivalently,
B
t
i(T t)B
t
= B
t
[1 i(T t)[ = B
T
(2)
The short-term spot interest rate is thus seen as one that determines the
interest due on the loan in the following way: The interest due, i(T t)B
t
,
is directly proportional to the amount loaned B
t
and to the horizon of the
loan (T t). In other words, the interest to be paid for xed B
T
is a linear
function of the horizon (T t).
1.1.2 Horizons longer than one year
We will consider two cases: First, the horizon is an integer number of
years; second, the horizon is not an integer (for instance, T t = 3X45
years).
1.1.2.1 Horizons longer than one year and integer numbers
For horizons longer than one year and integer numbers, another way of
determining and computing rates of interest is generally used. Since over a
number of years the amount borrowed often does not imply payment of
interest at the end of each year but only at the maturity of the loan, the
amount due at T, after T t years, is the following, applying the well-
known principle of compound interest:
B
t
(1 i)
Tt
= B
T
(3)
Therefore, the rate of interest i, or the rate of return on a loan starting at t
and reimbursed T t years later, is
i =
B
T
B
t
_ _
1a(Tt)
1 (4)
For instance, consider a zero-coupon bond with three years of maturity,
par value B
T
= 100, and value today B
t
= 78X48. Applying (4), the rate of
return on that default-free bond is 8.413%. And we can infer that the
three-year spot rate, or the rate of interest for three-year loans, is 8.413%.
1.1.2.2 Horizons longer than one year and fractional
We can tackle the case of a horizon T t larger than one year and frac-
tional in the following way. When horizons are longer than one year and
not an integer number (for instance, 3.45 years), there are two possibil-
ities: Either directly apply formula (4), replacing T t by 3.45, which
leads to 7.277%; or, logically, consider applying a combination of formula
(4) (compound interest), the integer number of years (3 years) and for-
mula (2) (simple interest), the remaining fraction of a year (0.45 of a
year). This implies that our rate of interest would be calculated as follows.
When T t designates the integer part of the horizon and : the remaining
fraction of the year (we have T t = T t :; in our example, T t =
3X45 years; T t = 3 years and : = 0X45 year), i would be such that
B
t
(1 i)
Tt
B
t
(1 i)
Tt
:i = B
t
(1 i)
Tt
(1 :i) = B
T
(5)
Unfortunately, equation (5) cannot be solved analytically for i, and
only trial and error (or numerical methods) can be used. Applying such
methods would lead to a value of i = 7X2576%.1 Since this method is
cumbersome, formula (4) is usually preferred.
There are many other ways to determine a rate of return or rate of
interest. First, an interest can be calculated and paid once a year, or a
number of times per year. (It can be paid, for instance, twice a year, four
times a year, every month; for reasons that will become clear later in this
book, dealt with in detail in chapter 11, it can even be considered to be
calculated an innite number of times per periodequivalently, it can
then be considered as being paid continuously.)
1.2 Forward rates and an introduction to arbitrage
On the other hand, as we mentioned earlier, interest rates can be agreed
upon today for loans starting only at a future date, for a xed horizon
or trading period; those are forward rates. Call f (0Y tY T) a forward rate
decided upon today, at time 0, for a loan starting at t and reimbursed at T.
A spot rate (for a loan that starts at time 0 and is reimbursed at any time
u) is a particular case of a forward rate. We denote it f (0Y 0Y u) Ii(0Y u)
for a maturity u, or f (0Y 0Y t) Ii(0Y t) for a maturity t. The arbitrage-
enforced links between forward rates and spot rates are very strong, and
we will briey explain here why.
There is in fact an arbitrage-driven equivalency either in borrowing or
in lending directly at rate i(0Y T)Y on the one hand, or via the shorter rate
1The reason we get a lower rate of return applying equation (5) instead of equation (4) is that
1 :i is larger than (1 i)
:
when 0 ` : ` 1; the converse is true for : b 1.
6 Chapter 1
i(0Y t) and the forward rate f (0Y tY T), on the other. This equivalency must
translate via this equation:
[1 i(0Y t)[
t
[1 f (0Y tY T)[
Tt
= [1 i(0Y T)[
T
(6)
because arbitrageurs would otherwise step in, and their very actions
would establish this equilibrium. We will now demonstrate this, but rst
we must dene what we mean by arbitrage.
Arbitrage consists of taking nancial positions that require no invest-
ment outlay and that guarantee a prot. At rst sight, it may seem strange
that a prot may be generated without any upfront personal outlay or
expenditure, but in many circumstances this is precisely the case. We will
now present such a case, supposing to that eect that the left-hand side of
(6) is larger than the right-hand side. We would thus have
[1 i(0Y t)[
t
[1 f (0Y tY T)[
Tt
b [1 i(0Y T)[
T
This means it is more rewarding to lend and more expensive to borrow
over a period T via two contracts (the spot, short one, and the forward
contract) than via the unique long-term contract. Arbitrageurs would im-
mediately borrow $1 on the long contract, invest this dollar in the short
contract, and at the same time invest the proceeds to be received at t,
[1 i(0Y t)[Y on the forward market for a time period T t. The forward
rate being f (0Y tY T), they would receive [1 i(0Y t)[
t
[1 f (0Y tY T)[
Tt
at
time T, which is larger than their disbursement [1 i(0Y T)[
T
by hypothe-
sis. Thus, without any initial outlay on their part, they would reap at time
T a sure prot equal to [1 i(0Y t)[
t
[1 f (0Y tY T)[
Tt
[1 i(0Y T)[
T
b 0.
What would be the consequences of such actions? First, demand would
increase on the long spot market, and the price of loanable funds on that
market, i(0Y T), would start to increase; second, supply both on the short
spot market and on the forward market would start to increase, entailing
a decrease in the prices of funds on those markets, i(0Y t) and f (0Y tY T),
respectively. This would take place until, barring transaction costs, equi-
librium was restored.
But arbitrageurs would not be the only ones to ensure a quick return to
equilibrium. Consider the `òrdinary'' operators on these three markets;
by `òrdinary'' operators we mean the persons, rms, or government
agencies that are the lenders or the borrowers acting on those markets as
simple investors or debtors; they simply carry out the operation for its
own sake. Suppliers, or lenders, have no reason to lend on the long spot
market when, at no additional risk, they can get higher rewards on the
short spot market and on the forward market; thus their supply will
shrink on the long spot market, contributing to the increase of i(0Y T).
Supply will therefore increase on the short spot market and on the for-
ward market, thus driving down i(0Y t) and f (0Y tY T). And of course,
symmetrically, borrowers have no reason to take the two-market route
when they can borrow directly on the long spot market; demand will
decline on the former markets and will increase on the latter, thus con-
tributing to the increase of i(0Y T) and the decrease of i(0Y t) and
f (0Y tY T).
One can easily see what would happen if initially the converse were
true; arbitrageurs, if they reacted quickly enough, could reap the (posi-
tive) dierence between the right-hand side of (6) and its left-hand side by
borrowing on the shorter markets and lending the same amount on the
long one. The rest of the analysis would be the same, and all results would
be symmetrical. We thus can conclude that, barring special transaction
costs, equation (6) will be maintained by forces generated both by arbi-
trageurs and ordinary operators on those nancial markets.
Equation (6) permits us to express the forward rate f (0Y tY T) as a
function of the spot rates, i(0Y t) and i(0Y T). Indeed, we can rearrange (6)
to obtain
f (0Y tY T) =
[1 i(0Y T)[
Ta(Tt)
[1 i(0Y t)[
ta(Tt)
1
(7)
Any interest rate plus one is usually called the ``dollar return''; it is equal
to the coecient of increase of a dollar when invested at the yearly rate
i(0Y t). For instance, if a rate of interest (spot or forward) is 7% per year,
1.07 is the dollar return corresponding to that rate of interest. We can
deduce from (6) an important property of the spot dollar return
1 i(0Y T). Taking (6) to the 1aT power, it can then be written as
1 i(0Y T) = [1 i(0Y t)[
taT
[1 f (0Y tY T)[
(Tt)aT
(8)
Remembering that the spot rate i(0Y t) can be considered as the forward
rate at time 0 for a loan starting immediately, written f (0Y 0Y t), we can see
immediately that the T-horizon spot dollar return, 1 i(0Y T), is the
8 Chapter 1
weighted geometric average2 of the dollar forward returns, the weights
being the shares of the various trading periods for the forward rates in the
total trading period T.
Of course, what we have demonstrated through arbitrage and tradi-
tional trading in equations (6), (7), and (8) generalizes to any arbitrary
partitioning of the time interval [0Y T[; let t
1
Y t
2
Y F F F Y t
n
be an arbitrary
partitioning of [0Y T[ into n intervals, those intervals being not necessarily
the same size. We then have the arbitrage-enforced equation
[1 i(0Y t
1
)[
t
1
[1 f (0Y t
1
Y t
2
)[
t
2
t
1
F F F [1 f (0Y t
n1
Y t
n
)[
t
n
t
n1
= [1 i(0Y t
n
)[
t
n
(9)
The t
n
horizon spot dollar return is the geometric average of all dollar
forward rates:
1 i(0Y t
n
) =
n
j=1
[1 f (0Y t
j1
Y t
j
)[
(t
j
t
j1
)at
n
(10)
Forward rates will be discussed again in chapter 9 of this book; we will
deal in more detail with the way interest rates are calculated in chapter 11.
Finally, forward rates will be used extensively in chapters 14 and 15.
1.3 Creating synthetically a forward contract
We will now show that we are always able to create synthetically a for-
ward contract from spot rates. Indeed, suppose you want to borrow $1 at
the future date t, to be reimbursed at T.
If you want to receive $1 at t, you can borrow on the long market an
amount (to be determined) and lend it on the short market; the amount to
be lent must be such that, at rate i(0Y t), it becomes 1 at t. It must there-
fore be equal to 1a[1 i(0Y t)[
t
. If this is the amount you borrow at time 0,
you owe 1a[1 i(0Y t)[
t
[1 i(0Y T)[
T
at time T. You have incurred no
disbursement at time 0, because at the same time you have borrowed and
lent 1a[1 i(0Y t)[
t
; at time t you have received 1, and at time T you re-
imburse the amount just mentioned; notice, of course, this corresponds
2Recall that the geometric weighted average of x
1
Y F F F Y x
n
with weights f
1
Y F F F Y f
n
(such that
n
j=1
f
j
= 1) is G =

n
j=1
x
fj
j
. We will make use of this concept again in chapter 13, when
dealing with the estimation of the long-term expected return of a bond or bond portfolios.
exactly to the amount we would have obtained applying the forward rate
dened implicitly in (6) and explicitly in (7). Indeed, applying that rate to
$1 borrowed on the forward market, we would have to repay at T
[1 f (0Y tY T)[
Tt
= 1
[1 i(0Y T)[
Ta(Tt)
[1 i(0Y t)[
ta(Tt)
1
_ _
Tt
=
[1 i(0Y T)[
T
[1 i(0Y t)[
t
(11)
which is exactly the amount to be reimbursed in this synthetic forward
contract made up of two spot rate contracts.
1.4 But why do interest rates exist in the rst place?
Until now we have examined a variety of interest rates, depending on the
date of the contract and the trading period. In particular, we were led to
distinguish between spot rates and forward rates. However, there is an
important question that we could have asked in the rst place: why do
interest rates exist? The answer to this question is not so obvious and
merits some comments.
Most people, when asked this apparently innocuous question, respond
that since ination is a pervasive phenomenon common to all places and
times, it is therefore only natural that lenders will want to protect them-
selves against rising prices; borrowers might well agree with that, there-
fore establishing the existence of an equilibrium rate of interest.
It is rather strange that ination is so often the rst explanation given
for the existence of interest rates. In fact, positive interest rates have
always been observed in countries where there was practically no ina-
tionor even a slight decrease in prices. Although ination is certainly a
contributor to the magnitude of interest rates, it is in no way their primary
determinant. There are two other principal causes.
First, and foremost, an interest rate exists because one monetary unit
can be transformed, through xed capital, into more than one unit after a
certain period of time. Some people are able, and willing, to perform that
rewarding transformation of one monetary unit into some xed capital
good. On the other hand, some are willing to lend the necessary amount,
and it is only tting that a price for that monetary unit be established
10 Chapter 1
between the borrowers and the lenders. Let us add that on any such
market both borrowers and lenders will nd a benet in this transaction.
Let us see why.
Suppose that at its equilibrium the interest rate sets itself at 6%. Gen-
erally, lots of borrowers would have accepted a transaction at 7%, or even
at much higher levels, depending on the productivity of the capital good
into which they would have transformed their loan. The dierence be-
tween the rate a borrower would have accepted and the rate established in
the market is a benet to him. On the other hand, consider the lenders:
they lent at 6%, but lots of them might well have settled for a lower rate of
interest (perhaps 4.5%, for instance). Both borrowers and lenders derive
benets (what economists call surpluses) from the very existence of a
nancial market, in the same way that buyers and suppliers derive sur-
pluses from their trading at an equilibrium price on markets for ordinary
goods and services.
Second, an interest rate exists because people usually prefer to have a
consumption good now rather than later. For that they are prepared to
pay a price; and as before, others may consider they could well part with a
portion of their income or their wealth today if they are rewarded for
doing so. There again, borrowers and lenders will come to terms that will
be rewarding for both parties, in a way entirely similar to what we have
seen in the case of productive investments.
Together with expected ination, these reasons explain the existence of
interest rates. The level of interest rates will depend on the attitudes of the
parties regarding each kind of transaction, and of course on the perceived
risk of the transactions involved.
An important point should be made here that, to the best of our
knowledge, has received short shrift until now. Anyone can agree that, for
a certain category of loanable funds, there exists at any point in time a
supply and a demand that lead to an equilibrium price, that is, an equi-
librium rate of interest. We should add here that, as in any commodity
market, the mere existence of an equilibrium interest rate set at the inter-
section of the supply-and-demand curves for loanable funds entails two
major consequences: First, the surplus of society (the surplus of the
lenders and the borrowers) is maximized (shaded area in gure 1.1);
second, the amount of loanable funds that can be used by society is also
maximized (gure 1.1).
Suppose now that for some reason interest rates are xed at a level
dierent from the equilibrium level (i
e
). Two circumstances, at least, can
lead to such xed interest rates. First, government regulations could pre-
vent interest rates from rising above a certain level (for instance i
0
, below
i
e
). Alternatively, a monopoly or a cartel among lenders could x the
interest above its equilibrium value i
e
(for instance, at i
1
).
If interest rates are xed in such a way, the funds actually loaned will
always be the smaller of either the supply or the demand. Indeed, the very
existence of a market supposes that one cannot force lenders to lend
whatever amount the borrowers wish to borrow if interest rates are below
their equilibrium valueand vice versa if interest rates are above that
value. We can then draw the curve of eective loans when interest rates
are xedthe kinked, dark line in gure 1.2. Should the interest rate be
xed either at i
0
or at i
1
, the loanable funds will be l (less than l
e
), and the
loss in surplus for society will be the shaded area abE. Innocuous as these
properties of nancial markets may seem, they are either unknown or
ignored by governments when they nationalize banksor when they
close their eyes to cartels established among banks (nationalized or not).
E
D
S
Loanable funds l
e
Interest
rate
i
e
Borrowers
surplus
Lenders
surplus
Figure 1.1
An equilibrium rate of interest on any nancial market maximizes the sum of the lenders' and
the borrowers' surpluses, as well as the amount of loanable funds at society's disposal.
12 Chapter 1
1.5 Kinds of bonds, their quotations, and a rst approach to yields on bonds
There are two main ways to distinguish between bonds: First, according
to their maturity, starting with personal saving deposits or very short-
term deposits, going to long-term (for instance, 30-year maturity secu-
rities); and second, by separating bonds according to their level of
riskiness, from entirely risk-free bonds issued by the U.S. Treasury, for
example, to so-called ``junk'' bonds. Since this book is concerned mainly
with defaultless bonds, we will consider bonds according to their matu-
rity. But we will also say a word about ranking bonds according to their
riskiness.
1.5.1 Treasury bills
Treasury bills have a maturity of one year or less and of course are issued
on a discount basis; their par value is usually $10,000. In table 1.1 (see the
last part of the table), six features of Treasury bills are presented:
par = par value of the Treasury bill (for instance: par = $10,000)
E
a
b
D
S
Loanable funds l
e
l
Interest
rate
i
e
i
1
i
0
Figure 1.2
Any articially xed interest rate reduces both the amount of loanable funds at society's dis-
posal and the surplus of society (the hatched area).
Table 1.1
Information about U.S. Treasury Bonds, notes, and bills
U.S. TREASURY I SSUES
Monday, June 7, 1999
Representative and indicatve Over-the-Counter quotations based on transactions of $1 million or more.
Treasury bond, note and bill quotes are as of mid-afternoon. Colons in bond and note bid-and-asked quotes represent 32nds; 101:01
means 101 1/32. Net changes in 32nds. Treasury bill quotes in hundredths, quoted on terms of a rate of discount. Days to maturity cal-
culated from settlement date. All yields are based on a one-day settlement and calculated on the offer quote. Current 13-week and
26-week bills are boldfaced.
For bonds callable prior to maturity, yields are computed to the earliest call date for issues quoted above par and to the maturity.
date for issues quoted below par. n-Treasury note.wi-When issued; daily change is expressed in basis points.
Source: Dow Jones Telerate/Cantor Fitzgerald
`Ù.S. Treasury Issues, Monday, June 7, 1999,'' by Dow Jones Telerate/Cantor Fitzgerald,
copyright 1999 by Dow Jones & Co., Inc. Reprinted by permission of Dow Jones & Co.,
Inc. via Copyright Clearance Center.
d
b
= the bid discount rate applied by dealers when they buy
a Treasury bill
p
b
= the dealers' price bid when buying a Treasury bill,
resulting from the discount bid rate
n = the number of days remaining until the T-bill's maturity
360 = the number of days in a year arbitrarily chosen when linking
d
b
to p
b
, and when calculating the `àsk yield''
0 = na360 Ifraction of an `àccounting year'' remaining
until the T-bill's maturity
The formula linking the discount bid d
b
(quoted in nancial news-
papers) to the price oered by the dealers p
b
is then calculated in the
following natural way:
d
b
=
par p
b
par
_
0 =
par p
b
par

360
n
(12)
Conversely, p
b
can be calculated from (12) as
0 par d
b
= par p
b
Therefore,
p
b
= par(1 0d
b
) = par 1
n
360
d
b
_ _
(13)
In consequence, any quoted bid discount translates immediately into an
oered price p
b
by the dealer. For instance, consider the T-bill on the last
line of table 1.1. Its remaining maturity is 352 days, and its bid discount is
4.80% per accounting year. Therefore, applying (13), the price oered by
the dealers is
p
b
= $10Y000 1
352
360
(0X048)
_ _
= $9530X67
for a transaction of $1 million or more.
Let us now introduce the following additional notation:
d
a
= the asked discount rate applied by dealers
when the dealers sell a Treasury bill
p
a
= the dealers' asked price when they sell a Treasury bill
Of course, the asked discount rate d
a
is related to the dealers asked price
p
a
by a formula entirely symmetrical to (1):
d
a
=
par p
a
par
_
0 =
par p
a
par

360
n
(14)
Conversely, the asked price is linked to the quoted asked discount by
p
a
= par(1 0d
a
) = par 1
n
360
d
a
_ _
(15)
Let us consider the same Treasury bill as before. Its asked discount is
4.79% per accounting year; therefore, the price asked by the dealers is
p
a
= $10Y000 1
352
360
(0X0479)
_ _
= $9531X64
Suppose now that we want to know the yield y received by the buyer if
he keeps his Treasury bill until maturity. He will of course make a calcu-
lation with a 365-day year. His yield will be
y =
par p
a
p
a
_
n
365
=
par
p
a
1
_ _
365
n
In our example, the yield received by the bond's buyer is
y =
10Y000
9531X64
1
_ _
365
352
= 5X09%ayear
However, the `àsk yield'' printed in the Wall Street Journal (5.03%)
still takes into account the `àccounting'' year of 360 days, and not the
calender year of 365 days. This is why they get
y
a
Iprinted asked yield =
10Y000
9531X64
1
_ _
360
352
= 5X03% per accounting year
It is easy enough to see why the yield for the buyer will always be
higher than the asked discount. We can see that
y b d
a
16 Chapter 1
since the above inequality translates as
par p
a
p
a
365
n
b
par p
a
par

360
n
This inequality leads to
365
p
a
b
360
par
which is always true, since p
a
` par. In fact, the ratio yield/asked dis-
count is equal to
y
d
a
=
par
p
a
365
360
b 1
a product of two numbers that are both larger than one. In our example,
this ratio is 1.0637.
1.5.2 Treasury notes
These are bonds with maturities between one and 10 years; they could be
called ``medium-term'' bonds. They usually pay semi-annual coupons in
the United States, contrary to many European bonds with the same
maturity that pay only one coupon yearly.
For those coupon-bearing, longer-term (more than a year) bonds, the
quotations are made on quite a dierent basis than the Treasury bills;
instead of quoting a bid or asked discount, professionals, together with
the nancial newspapers, quote both a bid price and an asked price (ask
price). In turn, those prices have two main features: First, they are not
quoted in dollars, but in percent of par value; second, the gure following
those percents is not a decimal but
1
32
. For instance, a Treasury note with
given maturity (between 1 and 10 years) and an ask price of 105.17 means
that the ask price is 105
17
32
= 105X53125% of its nominal value (
1
32
trans-
lates into 3.125% of one percent3).
The ask yield of that bond requires that we explain in detail what a
bond's yield to maturity is. We do this in chapter 3, where the central
concepts of yield to maturity and horizon rate of return are discussed.
3In nancial parlance,
1
100

1
100
=
1
10Y000
= one ten-thousandth and is called `òne basis point'';
so
1
32
of 1% is 3.125 basis points. For instance, if an interest rate, initially at 4%, increases by
50 basis points, it means that it has gone from 0.04 to 0.045, or from 4% to 4.5%.
However, for those who are already familiar with the concept of the
internal rate of return on an investment (IRR), suce it to say that the
bond's ask yield exactly corresponds to the IRR of the investment con-
sisting of buying this bond, with one proviso: the price to be paid by the
investor is not just the ask price; the buyer has to pay the ask price plus
whatever accrued interest has incurred on this bond that has not yet been
paid by the issuer of the debt security. Consider, for instance, that an
investor decides to buy a bond with an 8% coupon. As is often the case in
the United States, this bond pays the coupon semi-annually, that is, 4%
of the par value every six months. Suppose three months have elapsed
since the last semi-annual coupon payment. The holder of the bond will
receive the ask price, plus the accrued interest, which is 2% of the bond's
par value. So the cost to the investor, which is the price he will have to
pay and consider in the calculation of his internal rate of return, is the
quoted ask price (translated in dollars by multiplying the ask price by the
par value), plus the incrued interest.4 This is the cost the investor will
have to take into account for the calculation of his internal rate of return,
or, in bond parlance, its ask yield to maturity.
1.5.3 Treasury bonds
These bonds are long-term: their maturity is longer than 10 years. Some
of them have an embedded feature, in the sense that they are ``callable'';
they can be reimbursed at a date that is usually xed between 5 and 10
years before maturity. For the Treasury, this feature enables it to take
into account a drop in interest rates in order to reissue bonds with smaller
coupons.
Table 1.1, extracted from the Wall Street Journal, summarizes much of
this information regarding the quotation of bonds.
1.6 Bonds issued by rms
The bonds issued by rms, also called corporate bonds, generally have
characteristics very close to those described abovein particular, they
might be callable by the issuer. Their trademark, of course, is that they
4The quoted price is also called the at price. The quoted, or at, price plus incrued interest
is called the full, or invoice price. In England, British government securities are called gilts;
their prices carry special names: the quoted, or at, price is called ``clean price''; the full, or
invoice, price is the ``dirty price.''
18 Chapter 1
are usually more risky than Treasury bonds. Their degree of riskiness is
evaluated by two large organizations: Standard & Poor's and Moody's. In
table 1.2 the reader will nd Standard & Poor's debt rating denition;
table 1.3 gives Moody's corporate rankings.
1.7 Convertible bonds
Convertible bonds are hybrid nancial instruments that lie between
straight bonds and stocks. They are bonds that give their holder the right
(but not the obligation) to convert the bond (issued generally by a com-
pany) into the company's shares at a predetermined ratio. This ratio (the
conversion ratio) r is the number of shares to which one bond entitles
its owner. For instance, in January 1999, an on-line bookseller issued a
10-year convertible with a 4.75% coupon. Each $1000 bond could be
converted into 6.408 shares. (For the story of that issue, see the excellent
article by Mitchell Martin in the New York Herald Tribune of February
1314, 1999.) Suppose that the price at which the convertible bond was
issued was exactly par; it then follows that its holder would have converted
Table 1.2
Standard & Poor's debt rating denitions
AAA Debt rated `ÀAA'' has the highest rating assigned by Standard & Poor's.
Capacity to pay interest and repay principal is extremely strong.
AA Debt rated `ÀA'' has a very strong capacity to pay interest and repay principal
and diers from the higher-rated issues only in small degree.
A Debt rated `À'' has a strong capacity to pay interest and repay principal
although it is somewhat more susceptible to the adverse eects of changes in
curcumstances and economic conditions than debt in higher-rated categories.
BBB Debt rated ``BBB'' is regarded as having an adequate capacity to pay interest
and repay principal. Whereas it normally exhibits adequate protection
parameters, adverse economic conditions or changing circumstances are more
likely to lead to a weakened capacity to pay interest and repay principal for debt
in this category than in higher-rated categories.
BB, B,
CCC,
CC, C
Debt rated ``BB,'' ``B,'' ``CCC,'' ``CC,'' and ``C'' is regarded, on balance, as
predominantly speculative with respect to capacity to pay interest and repay
principal in accordance with the terms of the obligation. ``BB'' indicates the
lowest degree of speculation and ``C'' the highest degree of speculation. While
such debt will likely have some quality and protective characteristics, these are
outweighed by large uncertainties or major risk exposures to adverse conditions.
CI The rating CI is reserved for income bonds on which no interest is being paid.
D Debt rated D is in payment default.
Source: Standard & Poor's Bond Guide, June 1992, p. 10.
Table 1.3
Moody's corporate bond ratings
Aaa
Bonds that are rated Aaa are judged to be of the best quality. They carry the smallest
degree of investment risk and are generally referred to as ``gilt edge.'' Interest payments are
protected by a large or exceptionally stable margin, and principal is secure. While the
various protective elements are likely to change, the changes that can be visualized are
most unlikely to impair the fundamentally strong position of such issues.
Aa
Bonds that are rated Aa are judged to be of high quality by all standards. Together with
the Aaa group they comprise what are generally known as high-grade bonds. They are
rated lower than the best bonds because margins of protection may not be as large as in
Aaa securities; or the uctuation of protective elements may be of greater amplitude; or
other elements may be present that make the long-term risks appear somewhat larger than
in Aaa securities.
A
Bonds that are rated A possess many favorable investment attributes and are considered to
be upper medium-grade obligations. Factors giving security to principal and interest are
considered adequate but elements may be present which suggest a susceptibility to
impairment sometime in the future.
Baa
Bonds that are rated Baa are considered to be medium-grade obligations, i.e., they are
neither highly protected nor poorly secured. Interest payments and principal security
appear adequate for the present but certain protective elements may be lacking or may be
characteristically unreliable over any great length of time. Such bonds lack outstanding
investment characteristics and in fact have speculative characteristics as well.
Ba
Bonds that are rated Ba are judged to have speculative elements; their future cannot be
considered as well assured. Often the protection of interest and principal payments may be
very moderate and thereby not well safeguarded during future good and bad times.
Uncertainty of position characterizes bonds in this class.
B
Bonds that are rated B generally lack characteristics of the desired investment. Assurance
of interest, principal payments, or maintenance of other terms of the contract over any long
period of time may be small.
Caa
Bonds that are rated Caa are of poor standing. They may be in default or elements of
danger may be present with respect to principal or interest.
Ca
Bonds that are rated Ca represent obligations that are highly speculative. They are often in
default or have other marked shortcomings.
20 Chapter 1
it into shares if the share price had been more than the par/conversion
ratio = $1000a6X408 = $156X05. Of course, the price of the stock that day
was lower than that benchmark: it was $122.875.
The possibility given to the owner of the bond to convert it into shares
has two drawbacks: rst, the coupon rate is always lower than the level
corresponding to the kind of risk the market attributes to the issuer;
second, convertibles rank behind straight bonds in case of bankruptcy of
the rm.
In order to get a clear picture of the option value embedded in the
convertible, let us consider that a rm has S shares outstanding and has
issued C convertible bonds with face value B
T
. Suppose also that the
bonds are convertible only at maturity T. Let us rst determine what
share (z) of the rm the convertible bond holders will own if they convert
at maturity. This share will be equal to the number of securities they were
allowed to convert for their C convertibles, rC, divided by the total
number of securities outstanding after the conversion: S rC. So our
bondholders have access to the following fraction of the rm:
z =
rC
S rC
At maturity, suppose that the rm's value is V and that V is above the
par value of all convertibles issued, CB
T
. Convertible bonds' holders have
two choices: either they convert, or they do not. If they do convert, their
portfolio of stocks is worth zV (the share z of the rm they own times the
rm's value); if they do not convert, they will receive from the rm the par
value of their bonds, that is, CB
T
. So the decision to convert will hinge on
the rm's value at maturity. Conversion will occur if and only if
zV b CB
T
or
V b
CB
T
z
and the value of the convertible portfolio will be a linear function of the
rm's value, zV. If, on the other hand, zV ` CB
T
, conversion will not
take place; convertible holders will stick to their bonds and will receive a
value equal to par (B
T
) times the number of convertible bonds (C), CB
T
.
This quantity is independent from the rm's value.
Finally, suppose that the rm's value at maturity is less than CB
T
. In
that unfortunate case, as sole creditors of the rm the convertible holders
are entitled to the rm's entire value, that is, they will receive V.
Table 1.4 summarizes the convertible's portfolio value and the payo
to an individual convertible as functions of the rm's value.
So the convertible's value at maturity is just its par value if the rm's
value is between CB
T
and CB
T
az. If the rm's value is outside those
bounds, the convertible's value is a linear function of the rm's value:
VaC if V ` CB
T
and zVaC if V b
cB
T
z
. These functions are depicted in
gure 1.3(a). In gure 1.3(b), the option (warrant) part of the convertible
is separated from its straight bond part. We can write:
.
value of straight bond: min (
V
C
Y B
T
)
.
value of warrant: min (0Y
zV
C
B
T
)
This last value is the value of a call on the part of the rm with an exercise
price (per convertible) of B
T
az.
Most convertible bonds have more complicated features than the one
depicted above. In particular, conversion can be performed before matu-
rity. Furthermore, often the rm can call back the convertible, and the
holder can as well put it to the issuer under specied conditions. Thus a
convertible has a number of imbedded options: a warrant, a written (sold)
call feature, and a put feature.
The rst valuation studies of convertibles were carried out by Brennan
and Schwartz5 and Ingersoll6 in the case where interest rates are non-
Table 1.4
Convertible's value at maturity T
Firm's value Convertible's portfolio value Convertible's value
V ` CB
T
V VaC
CB
T
` V `
CB
T
z
CB
T
B
T
V b
CB
T
z
zV zVaC
5M. J. Brennan and E. S. Schwartz, ``The Case for Convertibles,'' Journal of Applied Cor-
porate Finance, 1 (1977): 5564.
6J. E. Ingersoll, `Àn Examination of Convertible Securities,'' Journal of Financial
Economics, 4 (1977): 289322.
22 Chapter 1
V
Firms value
Convertibles
value
B
T
V
C
V
C
CB
T
CB
T
(a)
V
Firms value
Convertibles
value
B
T
B
T
V
C
CB
T
CB
T
Straight bond
Warrant
(b)
Figure 1.3
(a) A convertible's value at maturity if stocks and convertibles are the sole nancial components
of its nancial structure. (b) Splitting the convertible's value at maturity into its components:
straight bond and warrant.
stochastic. The extension to a stochastic term structure was carried out by
Brennan and Schwartz7.
Questions
1.1. Suppose a default-free zero-coupon bond expires in four months
and is worth $97 today. What is the annualized return for its owner at
maturity?
1.2. A zero-coupon bond has a 6% annualized return per year and will be
redeemed in six months at $100. Supposing it is default-free, what is its
price today?
1.3. Suppose a zero-coupon bond has a maturity equal to 2.65 years;
its par value is $100, and its value today is $87. What is its annualized
return?
1.4. What is the price today of a zero-coupon bond that will be redeemed
at $1000 in 3.2 years, if its annualized rate of return is 6%?
1.5. Consider a two-year spot rate of 5% and a ve-year spot rate of 6%.
What is the (equilibrium) implied forward rate for a loan starting in two
years and maturing three years later? What is the interpretation of the
ve-year dollar spot return in terms of the two-year dollar spot return and
the dollar forward rate corresponding to the forward return you have just
calculated?
1.6. Suppose that today a six-year spot rate is 7% per year, and a forward
rate starting in four years and maturing two years later is 7.5% per year.
What is the implied four-year spot rate?
7M. J. Brennan and E. S. Schwartz, `Ànalyzing Convertible Bonds,'' Journal of Financial
and Quantitative Analysis, 15 (1980): 907929.
24 Chapter 1
2
AN ARBITRAGE-ENFORCED VALUATION
OF BONDS
Lady Macbeth. Thy letters have transported me beyond
This ignorant present, and I feel now
The future in the instant.
Macbeth
Chapter outline
2.1 An arbitrage-driven valuation of a zero-coupon bond 26
2.1.1 The case of an undervalued zero-coupon bond 27
2.1.2 The case of an overvalued bond 28
2.2 An arbitrage-enforced valuation of a coupon-bearing bond 29
2.2.1 The case of an undervalued coupon-bearing bond 30
2.2.2 The case of an overvalued coupon-bearing bond 32
2.3 Discount factors as prices 34
2.4 Evaluating a bond: a rst example 36
2.5 Open- and closed-form expressions for bonds 39
2.6 Understanding the relative change in the price of a bond through
arbitrage on interest rates 44
2.7 A global picture of a bond's value 52
2.8 How can we really price bonds? 54
Overview
In this chapter we evaluate a defaultless bond as a sum of cash ows to be
received in the future. Those future cash ows can be discounted in two
ways: either by assuming that all spot interest rates are the same, what-
ever the maturitywhich is very rarely the caseor by using for each
maturity a well-dened spot rate. Each method will provide a theoretical
value of the bond that may be compared with the market price; in that
sense it is possible to estimate a possible overvaluation or undervaluation
of the defaultless bond.
A defaultless bond may see its value change for two reasons. By far the
more important one is a change in rates of interest; a secondary one is the
passage of time. When interest rates move up, a bond's price goes down,
and vice versa. Those movements can be quite large and dicult to pre-
dict, which makes bond portfolio management both a rewarding and dif-
cult task. On the other hand, should a bond not be priced at par, the
bond price will tend toward par after some time has elapsed, until it
reaches par at maturity. This result can be derived from the fact that a
bond's value, as a percentage of its par value, can be shown to be the
weighted average between the coupon rate divided by the rate of interest
on the one hand, and one, on the other. The rst weight decreases through
time, which implies that the second weight increases through time, so that
the bond's price converges toward 100%.
Dene the direct yield of a bond as the ratio of the coupon to the price
of the bond. The capital gain on the bond in percentage terms will always
be the dierence between the rate of interest and the direct yield. This
result stems from arbitrage on interest rates, because in the absence of any
movement in interest rates, no one would ever buy a bond whose total
return (direct yield plus capital gain) would not equal the current rate of
interest.
It has become commonplace to say the present value of a future (cer-
tain) cash ow is the size of that cash ow discounted appropriately by a
discount factor. However, familiarity with that concept tends to mask the
true reason for that equation, which is the absence of arbitrage oppor-
tunities. Our intent in this chapter is to show how the future cash ows of
a bond can be valued today through arbitrage mechanisms, and how the
forces driven by arbitrage will lead to equilibrium prices. We will start
with the valuation of a zero-coupon bond. We will then consider the
valuation of a coupon-bearing bond, which is nothing other than a port-
folio of zero-coupon bonds, since it rewards its owner with a stream of
cash ows at precise dates.
2.1 An arbitrage-driven valuation of a zero-coupon bond
Consider that today, at time 0, the T-horizon spot rate is i(0Y T) and that
a zero-coupon bond pays at maturity T the par value B
T
. Is there a way
to determine its arbitrage-enforced value today? If a market exists for
rates of interest with maturity T on which it is possible to lend or borrow
at rate i(0Y T), there is denitely a way to nd an arbitrage-driven equi-
librium price for this zero-coupon bond. We will now show that any price
other than B
0
= B
T
[1 i(0Y T)[
T
would trigger arbitrage, which would
drive the bond's price toward B
0
. We will consider two cases: First, the
zero-coupon bond is undervalued with respect to B
0
= B
T
[1 i(0Y T)[
T
;
second, it is overvalued.
26 Chapter 2
2.1.1 The case of an undervalued zero-coupon bond
Suppose rst that the market value of the bond, denoted V
0
, is below B
0
;
we have the inequality
V
0
` B
0
= B
T
[1 i(0Y T)[
T
(1)
Let us show why this bond is undervalued. Suppose you borrow, at time
0, V
0
at rate i(0Y T) for period T; you will use the loan to buy the bond. At
time 0, your net cash ow is zero. Consider now your position at time T.
You owe V
0
[1 i(0Y T)[
T
, and you receive B
T
. Observe that (1) implies
also
B
T
b V
0
[1 i(0Y T)[
T
(2)
Inequality (2) implies that B
T
(what you receive at T) is larger than what
you owe, V
0
[1 i(0Y T)[
T
. Your net prot at T is thus the positive net
cash ow B
T
V
0
[1 i(0Y T)[
T
. Table 2.1 summarizes these transactions
and their related cash ows.
The important question to ask now is: what will happen on the nan-
cial market if such transactions are carried out? There is of course no
reason why only one arbitrageur would be ready to cash in a positive
amount at time T with no initial nancial outlay. Many individuals would
presumably be willing to play such a rewarding part. The result of such an
arbitrage would be to increase the value V
0
of the zero-coupon bond at
time 0 (since there will be excess demand for it), and to increase as well the
interest rate of the market of loanable funds with maturity T, because in
order to perform their operations, arbitrageurs will of course set those
markets in excess demand. Furthermore, anybody who had the idea of
lending on that market will prefer buying the bond, and all those who
Table 2.1
Arbitrage in the case of an undervalued zero-coupon bond; V
0
HB
0
FB
T
[1Bi(0, T)]
CT
Time 0 Time T
Transaction Cash ow Transaction Cash ow
Borrow V
0
at rate
i(0Y T) V
0
Reimburse loan V
0
[1 i(0Y T)[
Buy bond V
0
Receive par value of
zero-coupon bond B
T
Net cash ow: 0 Net cash ow:
B
T
V
0
[1 i(0Y T)[
T
b 0
27 An Arbitrage-Enforced Valuation of Bonds
wanted to issue bonds on that market will prefer borrowing instead. This
will contribute to pushing price V
0
and interest rate i
0
further up. These
moves will contribute to transforming inequality (1) (equivalently, in-
equality (2)) into an equality, barring any transaction costs that prevent
an equality from being established.
2.1.2 The case of an overvalued bond
It is useful to consider in some detail the opposite case of an overvalued
bond, because it is not entirely symmetrical to the case that has just been
treated. Suppose indeed that we now start with the inequality
V
0
b B
0
= B
T
[1 i(0Y T)[
T
(3)
Our rst idea, of course, is if an arbitrageur should buy an undervalued
security (as he has just done in our rst case), he should sell an overvalued
security, as is the case now. But how can he sell a security he does not
own? (Remember that if he initially owns the security, and if he sells it
because it is overvalued, he is not doing arbitrage; by denition, arbitrage
means earning something without investing any of one's own resources.
Since he is putting his money up front, he is investing his own resources.)
This is the point that makes the two cases asymmetrical: the arbitrageur
will have to take one step more than in the rst case. First, the arbitrageur
has to borrow the zero-coupon bond; this does not involve any cash ows.
He will then sell it for V
0
(bB
0
= B
T
[1 i(0Y T)[
T
) and invest the pro-
ceeds at rate i(0Y T) for period T. At T, he will receive V
0
[1 i(0Y T)[
T
.
Now (3) can also be written
V
0
[1 i(0Y T)
T
[ b B
T
(4)
Table 2.2
Arbitrage in the case of an overvalued zero-coupon bond; V
0
IB
0
FB
T
[1Bi(0, T)]
CT
Time 0 Time T
Borrow bond 0 Receive proceeds of
loan V
0
[1 i(0Y T)[
T
Sell bond at V
0
V
0
Buy back bond B
T
Invest proceeds of
bond's selling V
0
Give back bond to
its owner 0
Net cash ow: 0 Net cash ow:
V
0
[1 i(0Y T)
T
[ B
T
b 0
28 Chapter 2
So, with his earnings V
0
[1 i(0Y T)[
T
at time T, he can buy the bond for
its par value B
T
, give it back to the person he borrowed it from, and make
a prot V
0
[1 i(0Y T)
T
[ B
T
. Table 2.2 recapitulates these transactions
and their associated cash ows.
Of course, if the operations are not entirely symmetrical to those of the
rst case (one operation more is needed at time 0 and at time T: borrow-
ing the asset at time 0 and giving it back at T), the eect on the markets
are exactly opposite to those we described earlier: the bond's price V
0
will
be driven down by excess supply, and the interest rate i(0Y T) will also be
reduced by excess supply, until inequalities (3) and (4) are restored to
equalities.
We conclude from this that the present value of the zero-coupon bond
with par value B
T
is indeed an arbitrage-enforced value. Should any dif-
ference prevail between the market value of the bond at time 0, V
0
, and
B
T
[1 i(0Y T)[
T
, arbitrage would drive the value toward its equilibrium
value. We observe also that if this is true, then a zero-coupon bond value
and a rate of interest are two dierent ways of representing the same
concept: the time value of a monetary unit between time 0 and time T.
2.2 An arbitrage-enforced valuation of a coupon-bearing bond
We now turn to valuing a portfolio of zero-coupon bonds in the form of a
single coupon-bearing bond. However, our analysis could be immediately
extended to a portfolio of bonds.
Suppose there are T markets of loanable funds, each establishing a
price that is the spot interest rate i(0Y t); t = 1Y F F F Y T. Let c
t
designate the
bond's cash ow at time t; for t = 1Y F F F Y T 1Y c
t
is simply the coupon, c;
for t = T, the (last) cash ow is the sum of the last coupon paid, c, and
the bond's par value B
T
. We will now show that the equilibrium,
arbitrage-enforced value of the bond B
0
is the sum of all its discounted
cash ows, the rates of discount being the rates of interest i(0Y t), t =
1Y F F F Y T:
B
0
=
T
t=1
c
t
[1 i(0Y t)[
t
(5)
As before, we will distinguish between the case of a bond with an initial
value V
0
below B
0
, and the case of an overvalued bond.
2.2.1 The case of an undervalued coupon-bearing bond
Suppose we have the inequality
V
0
` B
0
=
T
t=1
c
t
[1 i(0Y t)[
t
(6)
Since we know both V
0
and B
0
, we can also determine a coecient
: (0 ` : ` 1) such that V
0
= :B
0
; : is thus the ratio of the bond's market
value V
0
to its arbitrage-enforced value B
0
(: = V
0
aB
0
` 1). We will now
show how an arbitrageur can prot from this situation.
She will rst borrow the amount V
0
= :
T
t=1
c
t
[1 i(0Y t)[
t
=
T
t=1
:c
t
[1 i(0Y t)[
t
, splitting this loan into the T loanable funds mar-
kets. On the rst market (the market with maturity of one year), she will
borrow :c
1
[1 i(0Y 1)[
1
; on the 2-year maturity market, she will bor-
row :c
2
[1 i(0Y 2)[
2
, and so forth. On the t-year market, her loan will be
:c
t
[1 i(0Y t)[
t
; on the T-year market, she will borrow : times the present
value of the cash ow she will receive at that time (which is c
T
, the last
coupon plus the par value B
T
), so she will borrow :c
T
[1 i(0Y T)[
T
. Her
net cash ow at time 0 is nil, since she has borrowed the amount V
0
,
which covers the price of the bond she has to pay.
After one year, she will have to reimburse :c
1
and will receive c
1
; after
the tth year, she will have to reimburse :c
t
and will cash in c
t
, and so
forth. At the bond's maturity, she will disburse :c
T
and receive c
T
. In all,
she will receive a series of positive net cash ows each year, equal to
c
t
:c
t
= (1 :)c
t
, t = 1Y F F F Y T, without having committed a single cent
of his own money in the process. The stream of the net cash ows
(1 :)c
t
, t = 1Y F F F Y T can be evaluated in terms of time 0 (in present
value), in terms of time T (in future value), or in terms of any year t for
that matter. For instance, let us evaluate it in present value. It will be
equal to
Net cash flow in present value =
T
t=1
(1 :)c
t
[1 i(0Y t)[
t
= (1 :)
T
t=1
c
t
[1 i(0Y t)[
t
(7)
30 Chapter 2
From
B
0
=
T
t=1
c
t
[1 i(0Y t)[
t
(8)
and V
0
= :B
0
, the arbitrageur's net cash ow is thus
(1 :)
T
t=1
c
t
[1 i(0Y t)[
t
= B
0
:B
0
= B
0
V
0
b 0 (9)
Our arbitrageur has earned, throughout the period [0Y T[, a net prot
equal, in present value, to exactly the dierence between the theoretical
value of the bond (B
0
) and its market value (V
0
). Table 2.3 summarizes
these transactions and their associated cash ows.
We now need to demonstrate that the theoretical value of the bond is
indeed the equilibrium, arbitrage-enforced value of the bond; that is, the
prices V
0
and B
0
will be driven toward each other by arbitrage forces.
Table 2.3
Arbitrage in the case of an undervalued coupon-bearing bond; V
0
HB
0
F
T
tF1
c
t
[1Bi(0, t)]
t
;
V
0
FB
0
, 0HH1
Time Transaction Cash ow
0 Borrow, on each of the t-year markets
(t = 1Y F F F Y T), :c
t
[1 i(0Y t)[
t
Buy bond for V
0
V
0
= :
c
t
[1 i(0Y t)[
t
V
0
Net cash ow: 0
1 Pay back :c
1
Receive rst cash ow (c
1
)
:c
1
c
1
Net cash ow: (1 :)c
1

t Pay back :c
t
Receive tth cash ow (c
t
)
:c
t
c
t
Net cash ow: (1 :)c
t

T Pay back :c
T
Receive Tth cash ow
:c
T
c
T
Net cash ow: (1 :)c
T
Total net cash ows in present value at time 0: (1 :)
T
t=1
c
t
[1 i(0Y t)[
t
= (1 :)B
0
=
B
0
:B
0
= B
0
V
0
b 0
This is easy to do, similar to what we have seen in the case of zero-
coupon bonds. First, when arbitrageurs are borrowing on each of the
t-term markets, they are creating an excess demand on each of those
markets, thus increasing each rate of interest i(0Y t), t = 1Y F F F Y n. This will
drive down the theoretical price of the bond, B
0
. Second, when buying the
bond, they are also creating an excess demand on the bond's market,
driving up its price V
0
. Barring transaction costs, this will take place until
V
0
= B
0
. We can of course add that if the magnitude of the transactions
on the bond is small because the existing stock of those bonds is small,
equilibrium will be reached only through the upward move of V
0
, not by a
signicant drop in B
0
, since the rates of interest i(0Y t) are not liable to
move upward in any signicant way. The same kind of remark would
have applied before when we considered the market for undervalued zero-
coupon bonds.
2.2.2 The case of an overvalued coupon-bearing bond
We will now consider how arbitrage will bring back an overvalued
coupon-bearing bond to its equilibrium value. We suppose here that
initially the market value of the bond is higher than its theoretical value
B
0
X V
0
b B
0
or V
0
= :B
0
, with : b 1. We have here
V
0
b B
0
=
T
t=1
c
t
[1 i(0Y t)[
t
or, equivalently,
V
0
= :B
0
= :
T
t=1
c
t
[1 i(0Y t)[
t
Y : b 1
This is how our arbitrageur will proceed. He will rst borrow the
overvalued bond, sell it for V
0
= :B
0
, and invest the proceeds in the
following way: On the one-year maturity market, he will invest
:c
1
[1 i(0Y 1)[
1
at rate i(0Y 1); on the t-year market, he will invest
:c
t
[1 i(0Y t)[
t
at rate i(0Y t), and so forth; on the T-year market, he will
invest :c
T
[1 i(0Y T)[
T
at rate i(0Y T).
Consider now what will happen on each of those terms t (t = 1Y F F F Y T).
At time 1, he will receive :c
1
from his loan. This is more than c
1
, since
: b 1, and he will have no problem in paying the rst coupon c
1
= c to
32 Chapter 2
the initial bond's owner; he will thus make at time 1 a prot :c
1
c
1
=
c
1
(: 1). The same operation will be done on each term t; at time t, he
will cash in :c
t
and pay out the bond's owner c
t
, making a prot c
t
(: 1).
At maturity our arbitrageur has two possibilities: either he buys back
the bond just before the issuer pays the last coupon c plus its par value
B
T
, and hands the bond to its initial owner (in which case his cash outow
is c
T
= (c B
T
)); or he can also pay directly the residual value of the
bond to its owner and incur the same cash outow of c
T
. On the other
hand, he has received :c
T
from the loan he made on the T-maturity
market. Whatever route he, or the bond's initial owner, may choose, our
arbitrageur will receive at maturity a positive cash ow c
T
:c
T
=
c
T
(1 :). Table 2.4 summarizes these operations and their related cash
ows.
Table 2.4
Arbitrage in the case of an overvalued coupon-bearing bond; V
0
IB
0
F
T
tF1
c
t
[1Bi(0, t)]
t
;
V
0
FB, I1
Time Transaction Cash ow
0 Borrow bond
Sell bond for V
0
Loan, on each t-term markets,
:c
t
[1 i(0Y t)[
t
Y t = 1Y F F F Y T
0
V
0
= :
T
t=1
c
t
[1 i(0Y t)[
t
V
0
Net cash ow: 0
1 Receive reimbursement on rst loan, :c
1
Pay rst coupon to bond's owner
:c
1
c
1
Net cash ow: (: 1)c
1
b 0

t Receive reimbursement on t-year loan, :c
t
Pay t-year coupon to bond's owner
:c
t
c
t
Net cash ow: (: 1)c
t
b 0

T Receive reimbursement on T-year loan,
:c
T
= :(c B
T
)
Pay last coupon plus par to bond's owner
:c
T
c
T
Net cash ow: (: 1)c
T
Total net cash ow in present value at time 0: (: 1)
T
t=1
c
t
[1 i(0Y t)[
t
= (: 1)B
0
=
:B
0
B
0
= B
0
V
0
b 0
Consider now the sum of all these positive cash ows in present value.
We have
prots Arbitrage in present value =
T
t=1
(: 1)c
t
[1 i(0Y t)[
t
= (: 1)
T
t=1
c
t
[1 i(0Y t)[
t
= (: 1)B
0
= :B
0
B
0
= V
0
B
0
b 0
The present value of the arbitrageur's prot will be exactly equal to the
overvaluation of the bond at time 0: V
0
B
0
. This result is exactly sym-
metrical to the one we reached previously: the arbitrageur could pocket
the precise amount of the undervaluation of the bond.
What will happen now in the nancial markets is quite obvious: Arbi-
trageurs will create an excess supply on the bond's market, thus driving
down the bond's price V
0
. On the other hand, the systematic lending by
arbitrageurs on each t-market will create an excess supply on those markets
as well, driving down all rates i(0Y t), t = 0Y F F F Y T, pushing up the theo-
retical price B
0
until V
0
reaches B
0
. Thus the equilibrium bond's value will
have been truly arbitrage-driven.
2.3 Discount factors as prices
An understanding of bond valuationand any investment for that
mattercan be gained through an economic interpretation of the dis-
count factors 1a(1 i
t
)
t
, t = 1Y F F F Y n as prices. Consider rst the cash ow
pertaining to the rst year, which we call c
1
. In present value terms, we
know that arbitrage arguments lead us to write this as c
1
a(1 i
1
), or
c
1
a(1 i) = c
1

1
1 i
1
Let us consider this present value of c
1
as a product, and let us examine
the units in which each term of the product is expressed. First, c
1
is in
monetary units of year one; for instance, it could be dollars of year one.
The ratio
1
1i
1
is in
$ of year zero
$ of year one
; so this is an exchange rate: $1 of year zero
34 Chapter 2
can be exchanged for $(1 i
1
) of year one. This is nothing but a price: the
price of $1 of year one expressed in dollars of year zero. Our present value
is therefore a number of units of dollars of year one (c
1
) times the price of
each dollar of year one in terms of dollars of year zero. We have
c
1
1 i
1
= c
1

1
1 i
1
= $ of year one
$ of year zero
$ of year one
= $ of year zero
Our discounting factor (1 i
1
)
1
really acts as a price, which serves to
translate dollars of year one into dollars of year zero.
This of course generalizes to any discount factor such as the ones we
have used in this chapter. The present value of cash ow c
t
, to be received
at time t, expressed in dollars of year zero, is
c
t
(1 i
t
)
t
= c
t

1
(1 i
t
)
t
As before, the right-hand side of this equation should be viewed as a
product of dollars of year t times the price of a dollar of year t expressed
in dollars of year zero
_
1
(1i
t
)
t
.
In terms of bonds, these prices (1 i
t
)
t
can be viewed as the arbitrage-
driven prices of zero-coupon bonds maturing in t years, with par value
equal to one dollar. Further in this text (rst in chapter 9), these prices of
zero-coupon bonds with par value 1 will become extremely useful. So we
will call them b(0Y t), and we will dene them as
1
(1 i
t
)
t
= b(0Y t)
Inversely, the spot interest rate i
t
can of course be expressed in terms
of the zero-coupon bond's price b(0Y t). From the above expression, we
derive
i
t
=
1
b(0Y t)
_ _
1at
1
An economic interpretation of this expression is interesting; in other
words, the answer can be derived just by reasoning, without recourse to
any calculation. For this purpose, the denition of gross return, also
called the dollar return, is useful. The gross, or dollar, return is the ratio
between what a dollar becomes after a given period and the dollar
invested; it is the principal (the dollar invested) plus interest paid, divided
by the dollar invested. This is exactly (1 i
t
)
t
, equal to 1ab(0Y t). The per
year gross return, or per year dollar return is the tth root of this, minus
one, that is, [(1 i
t
)
t
[
1at
1 = i
t
= [1ab(0Y t)[
1at
1. Hence no calcula-
tion is required to express the spot rate as a function of the zero-coupon
bond price.
2.4 Evaluating a bond: a rst example
Bonds promise to pay their holder coupons at regular intervals and to
reimburse the principal at a maturity date. As we have seen, their value is
none other than the present value of all cash ows that will be received by
their owner. We deal here mainly with a defaultless bond; this implies that
the nominal cash ows to be received can be considered completely risk-
free. We will rst evaluate a bond as a sum of cash ows and then indicate
formulations in open form whereby it will be possible to see at a glance
how a bond is a decreasing function of any interest rate, be it an interest
rate common to all maturities or a spot rate pertaining to any given
maturity. We will then indicate a closed-form formula that will prove
useful to understanding how the price of a bond will always ultimately
reach its par value at maturity.
The following question naturally arises: what kind of discounting
factor should be used to calculate the present value of any cash ow to be
received t years from now? Normally, we would use the riskless spot rate
for horizon t, as described below. Some simplied notation will be useful
here. Let us call
.
B
0
= the present value of the bond
.
B
T
= the face (or the reimbursement) value of a bond (example: $1000)
.
c = the annual coupon of the bond (example: $70)
.
caB
T
= the coupon rate of the bond (in this example, the coupon rate is
$70/$1000 = 7% per year)
.
i
t
= the spot rate prevailing today for a contract of length equal to t
years; i
t
represents the term structure of interest rates. For example, i
1
=
4X5%; i
2
= 4X75%; i
10
= 5X5%.
.
T = maturity of the bond; this is the number of years remaining until
maturity (example: 10 years).
36 Chapter 2
In summary, the bond we have considered in this example has a face
value of $1000; its coupon rate is 7% per year; it will be reimbursed in 10
years. Suppose that currently the one-year to 10-year spot rates exhibit a
slightly increasing structure: they range between 4.5% and 5.5%. Presum-
ably, since the coupon rate (7%) is signicantly above these spot rates, it
can be inferred that the bond was issued in earlier times, when interest rates
were higher. Now the present value of the bond is calculated as follows:
B =
c
1 i
1
c
(1 i
2
)
2

c
(1 i
t
)
t

c B
T
(1 i
T
)
T
(10)
Suppose that, for the 10 years remaining in the bond's life in our
example, the term structure is that shown in Table 2.5. This term structure
is supposed to be slightly increasing, leveling o for long maturities, in the
same way as has often been observed. We will show in chapter 9 exactly
how this spot rate term structure can be derived from a set of coupon-
bearing prices.
The present value of the bond, using the above term structure, would be
B =
70
1X045

70
1X0475
2

70
1X0495
3

70
1X051
4

70
1X052
5
70
1X053
6

70
1X054
7

70
1X0545
8

70
1X055
9

1070
1X055
10
= $1118X25
This value is above par, a result that conrms our intuition; since the rates
of interest corresponding to the various terms of the bond are below the
coupon rate, we expect the price of the bond to exceed its par value. Any
holder of this bond, which promises to pay a coupon rate above any of
the interest rates available on the market, will accept parting from his
bond only if he is paid a price above par; and any buyer will accept
paying at least a bit more than par to acquire an asset that will earn him a
return above the available interest rates.
Table 2.5
An example of a term structure
Term (years) 1 2 3 4 5 6 7 8 9 10
Spot rate (%) 4.5 4.75 4.95 5.1 5.2 5.3 5.4 5.45 5.5 5.5
This valuation process, therefore, constitutes a natural way to dene
the intrinsic value of a defaultless bond. Indeed, any price mismatch be-
tween the market value of the bond and its theoretical price would lead to
buying the asset, in the case of undervaluation by the market, and to
selling it, in the case of overvaluation. There is, however, one diculty,
which has to do with the estimation of the term structure of the rate of
interest. We generally have no direct knowledge of the set of spot rates
such as the one postulated above, because there are not enough nancial
markets that would have exactly the horizons required to construct such
a prole of spot rates. Indeed, one way of obtaining these rates would
be to consider a whole series of zero-coupon bonds, the maturity of which
would correspond to each term.
To illustrate, remember that a zero-coupon bond promises to pay a
single cash owthe face value (perhaps $1000)in t years. Let Z
0
be
the value today of that bond, and let F be the face value of the bond. To
determine the interest rate (the spot rate) i
t
implied by Z
0
, F, and the time
span t, i
t
must be such that it transforms Z
0
into F after t years, with
compounded interest. That means Z
0
(1 i
t
)
t
= F, which implies in turn
that the spot rate is simply i
t
= (FaZ
0
)
1at
1.
Unfortunately, we do not have the series of zero-coupon bonds
required for that purpose. We are then faced with the problem of evalu-
ating the term structure of interest rates. We will have more to say about
this later, but already we can discern that the value of a bond is inversely
related to any increase in this structure, because if every rate of interest
happened to increase, every cash ow provided by the bond would see
its present value diminish. Therefore, it may be rewarding to make the
simplication that the term structure is at (the spot rates are the same
whatever the horizon may be), so that we can infer the changes in the
theoretical price of a bond from a change in this single interest rate, which
we will choose to call i. If we do make this simplication, then i
t
= i for
all t = 1Y F F F Y T, and the evaluation formula (10) becomes
B(i) =
c
1 i

c
(1 i)
2

c
(1 i)
t

c B
T
(1 i)
T
(11)
It will be very useful to visualize the sum of these discounted cash ows.
In gure 2.1, the upper rectangles represent the cash ows of our bond in
current value, and the lower rectangles are their present (or discounted)
value. The value of the bond is simply the sum of these discounted cash
38 Chapter 2
ows. If the coupon rate is 7%, the interest rate 5%, and the maturity 10
years, the bond's value is $1154.
2.5 Open- and closed-form expressions for bonds
Sums like (11) are not expressed in a direct analytical form that would
allow us to infer its change in value from a change in the parameters it
contains. For instance, formula (11) establishes clearly that the bond's
value is a decreasing function of i, because it is a sum of functions that all
are decreasing functions of i; but what about the dependency between B
and T? Suppose that maturity T decreases by one year. Will B decrease,
increase, or stay constant? We cannot answer this question merely by
considering (11); indeed, if T diminishes by one unit and the par value
B
T1
= B
T
, the last fraction increases, but you have one coupon lessso
you cannot make a conclusion. It will therefore be useful to express (11) in
analytical form. It turns out that (11) simplies to a very convenient form,
as we shall demonstrate.
0
200
400
600
800
1000
1200
1 2 3 4 5
Years of payment
6 7 8 9 10
N
o
m
i
n
a
l

a
n
d

d
i
s
c
o
u
n
t
e
d

v
a
l
u
e
s

(
h
a
t
c
h
e
d

a
r
e
a
s
)
Figure 2.1
The value of a bond as the sum of discounted cash ows (sum of hatched areas; coupon: 7%;
interest rate: 5%).
From the value of a bond given by (11), set
c
1i
in factor; we get
B =
c
1 i
1
1
1 i

1
(1 i)
T1
_ _
B
T
(1 i)
T
(12)
The term in brackets is a geometrical series of the form 1 a
a
T1
, where a = 1a(1 i). Its sum1 is
1a
T
1a
, and therefore we have:
B =
c
1 i
1 (
1
1i
)
T
1
1
1i
_ _
B
T
(1 i)
T
=
c
i
1
1
(1 i)
T
_ _
B
T
(1 i)
T
(13)
Let us rst check that it has the right value at maturity (when the
remaining maturity T becomes zero). For T = 0, 1 1a(1 i)
T
is equal
to zero, and B
T
a(1 i)
T
becomes B
T
. So indeed the value of the bond is
at its par, as it should be.
Now look at what happens when there is no reimbursement (or,
equivalently, when maturity is innity). The bond, in this case, is called
a perpetual, or a consol; when T y, B tends toward cai. We then
have
Value of a perpetual: lim
Ty
B =
c
i
(14)
The reader may wonder whether formula (13), derived through some
simple calculations, has a direct economic interpretation: in other words,
is it possible to obtain it directly through some economic reasoning? The
answer is denitely positive if we keep in mind that the value of a per-
petual is cai. We are going to show that the price of any coupon-bearing
bond with nite maturity is in fact a portfolio of three bonds, two held
long and one held short. We can rst consider the value of the coupons
until maturity T: this is the dierence between a perpetual today (at time
0) (equal to
c
i
) and a perpetual whose coupons start being paid at time
T 1. Its value at time T is
c
i
, and its value today is just
c
i
(1 i)
T
, so the
present value of the coupons of the bond is a perpetual held long today,
1 Call S
T1
the sum S
T1
= 1 a a
2
a
T1
. We then can write aS
T1
= a a
2
a
3
a
T
. Now subtract aS
T1
from S
T1
: you get S
T1
aS
T1
= S
T1
(1 a) =
1 a
T
; therefore S
T1
= (1 a
T
)a(1 a).
40 Chapter 2
less a perpetual truncated from 0 to T held from T onward. The total
value of that portfolio today is
c
i

c
i
(1 i)
T
=
c
i
[1 (1 i)
T
[, and this
is the rst part of formula (13). Now we simply have to add the reim-
bursement in present value, B
T
(1 i)
T
, which may be considered as the
value of a zero-coupon, and we get our formula for the coupon-bearing
bond as a portfolio of a perpetual and a zero-coupon held long, less a
perpetual held at time T.
In practice, the bond's value is usually quoted per unit of par value B
T
.
Let us divide B by B
T
.
B
B
T
=
caB
T
i
1
1
(1 i)
T
_ _
1
(1 i)
T
(15)
This formula is very useful, because it can be immediately seen that the
value of a bond (relative to its par value) is simply the weighted average
between the coupon rate divided by the rate of interest, on the one hand,
and the number 1, on the other; the weights are just the coecients
1 (1 i)
T
and (1 i)
T
, which add up to 1, as they should.
Either formula(11) or (15)shows that the value of a defaultless
bond depends on two variables that are not in the control of its owner: the
rate of interest i and the remaining maturity T.
The advantage of formula (11) is that it can be seen immediately that
the value of the bond is a decreasing function of i. But (15) readily shows
two fundamental properties of bonds:
.
First, since the bond's value (relative to its par value) is always between
caB
T
i
and 1, the price of the bond will be above par if and only if the cou-
pon rate is above the rate of interest. Conversely, it will be below par if
and only if the coupon rate is below the rate of interest.
.
Second, when maturity T diminishes, the weight of 1that is,
1
(1i)
T
becomes larger and larger, so the value of the bond tends toward its par
value, to become equal to 1 when maturity T becomes zero. This process
will be accomplished by a decrease in value if initially the bond is above
par and by an increase in value if initially the bond is below par.
The evolution of our bond through time has been pictured in gure
2.2, for various rates of interest. Should the rate of interest change during
this process, the value of the bond would jump from one curve to
another. Suppose, for instance, that the interest rate is 5% and that the
time elapsed in our bond's life is four years. (Time to maturity is there-
fore six years.) Should interest rates drop to 3%, our bond's trajec-
tory would immediately jump up from 5% to 3%, and then follow that
path until maturityunless other changes also occur in the interest
rates.
Consider now the relationship between the rate of interest and the price
of the bond. From (11), as we have just said, we can see that the bond's
value is a decreasing function of the rate of interest. This relationship is
pictured for our bond in gure 2.3, where we consider various maturities.
(We have chosen 10, 7, 3, 1, and 0 years.) As could be foreseen from the
previous paragraph, the shorter the maturity, the closer the bond price is
to its par value. Also, for T = 0 (when the bond is at maturity), its price is
at par irrespective of the rate of interest. Naturally, any asset that is about
to be paid at its face value without default risk would have precisely this
0 2 4 6 8 10 12
700
1500
1400
1300
1200
1100
1000
900
800
Time (years)
V
a
l
u
e

o
f

t
h
e

b
o
n
d
3%
5%
7%
9%
11%
Figure 2.2
Evolution of the price of a bond through time for various interest rates (coupon: 7%; maturity:
10 years).
42 Chapter 2
value, whatever the interest rate may be. All curves, each corresponding
to a given maturity, exhibit a small convexity.2
This convexity is a fundamental attribute of a bond value and, more
generally, of bond portfolios. It is so important that we will devote an
entire chapter to it. Already we can immediately gure out that the more
convexity a bond exhibits, the more valuable it will become in case of a
drop in the interest rates, and the less severe will be the loss for its owner
if interest rates rise. We will also ask, in due course, what makes a bond,
or a portfolio, more convex than others.
In gure 2.2, we have already noticed that the value of a bond con-
verged toward its par value when maturity decreased. The natural ques-
2 From your high school days, you remember that a function's convexity is measured by its
second derivative. The function is convex if its second derivative is positive (if its slope is
increasing). This is the case in gure 2.3: each curve sees its slope increase when i increases
(its value decreases in absolute value but increases in algebraic value). A curve is concave if
its second derivative is negative. A curve is neither convex nor concave if its second deriva-
tive is zero; this implies that the rst derivative is a constant and therefore that the function is
represented by a straight line (an ane function).
2 3 4 5 6 7 8 9 10 11 12
700
1500
1400
1300
1200
1100
1000
900
800
Rate of interest (%)
V
a
l
u
e

o
f

t
h
e

b
o
n
d
10 yr
7 yr
3 yr
1 yr
0 yr
Figure 2.3
Value of a bond as a function of the rate of interest and of its maturity (coupon: 7%; maturities:
10, 7, 3, 1, and 0 year).
tion to ask now is: at what rate will the decrease or the increase in value
take place? This is an important question; if the owner of a bond takes it
for granted that for such a smart person as himself his asset gains value
through time, he will have some diculty accepting that his asset is
doomed to take a plunge (if only a small one) year after year when his
bond is above par.
The key to answering this question is to realize that, in the absence of
any change in the interest rate, the owner of a bond simply cannot receive
a return on his investment dierent from the prevailing interest rate.
Whenever the coupon rate is above the interest rate, the investor will see
the value of his bond, which he may have bought at B
0
, diminish one year
later to a value B
1
, such that the total return on the bond, dened by
caB
0
(B
1
B
0
)aB
0
, will be exactly equal to the rate of interest. We will
show this rigorously, but here is rst an intuitive explanation.
Suppose that the coupon c divided by the initial price B
0
(caB
0
) is
above the rate of interest, and that the value of the bond one year later
has not fallen to B
1
but slightly above it. The return to the bond's owner
is then above the rate of interest, and it is in the interest of the owner to
try to sell it at time 1. However, possible buyers will nd the bond over-
valued compared to the prospective future value of the bond (ultimately,
the bond is redeemed at par value), and this will drive down the price
of the bond until B
1
is reached. On the other hand, had the price fallen
below B
1
, owners of the bond would resist selling it, and possible buyers
would attempt to buy it, because it would be undervalued. Its price would
then go up, until it reached B
1
.
Let us now look at this matter a little closer; we will then be led to the
all-important equation of arbitrage on interest rate, which plays a central
role in nance and economics.
2.6 Understanding the relative change in the price of a bond through arbitrage on
interest rates
In the two rst sections of this chapter, we showed how the valuation of a
bond could be obtained through direct arbitrage. In this section, we want
to show how the change in price of a bond can be derived from arbitrage
on rates of interest. To that eect, we will derive the Fisher equation,
44 Chapter 2
using for that purpose two very dierent kinds of assets: a nancial asset
(a zero-coupon bond) and a capital goods asset in the form of a vineyard.
We have deliberately chosen two assets of an entirely dierent nature; by
doing so, we will clearly indicate the remarkable generality of Fisher's
equation. Also, we will postulate the absence of uncertainty. But how can
we escape uncertainty when considering a vineyard?
The answer is that we may transform a risky asset into a riskless one
by supposing the existence of forward markets. (Let us not forget that
forward and futures markets were createdthousands of years ago
precisely in order to remove uncertainty from a number of future trans-
actions.) In our case, we will suppose that an investor buys a vineyard (or
a piece of it), rents it for a given period, and resells it after that period;
thus, any uncertainty is removed from the outcomes of these transactions.
Let us also suppose that the rental fee is paid immediately, and that the
reselling of the vineyard is done in a forward contract with guarantees
given by the buyer such that all uncertainty is removed from the outcome
of that forward contract.
Let us rst ask the question, what can an investor do with $1? Suppose
he has two possibilities: invest it in a tangible asset (a vineyard, for
instance) or in a nancial asset. For the time being, suppose that both
assets are riskless. This means that the nancial asset (perhaps a zero-
coupon bond maturing in one year) will bring with certainty a return
equal to the interest rate i. It also means that the income that may be
generated by the vineyard, both in terms of net income and capital gains,
can be foreseen with certainty. This certainty hypothesis is much less of a
constraint than it may appear; we will suppress it later, but for the time
being we prefer to maintain this certainty for clarity's sake.
Let us take up the possibility of investing in the nancial market, at the
riskless rate. The owner of $1 will then receive after one year 1 i. Turn
now to the investor in the vineyard, and suppose that the price of this
vineyard today is p
0
. With $1, he can buy a piece
1
p
0
of it. (Of course, it is
of no concern to us that the asset is supposed to be divisible in such a
convenient way; should this not be the case, we would suppose that the
money at hand for the investor is $p
0
instead of $1, and he would then
compare the reward of investing p
0
in the nancial market, at the riskless
interest i, with buying the whole vineyard at price p
0
.)
Coming back to our investor, he now holds a piece
1
p
0
of the vineyard,
and he may very well want to rent it. Suppose that the yearly rent he can
nd for such a vineyard is q (the rental rate q is expressed in dollars per
vineyard, per year). After one year, he will receive a rental of q
1
p
0
=
q
p
0
.
The price of the vineyard will be p
1
at that time, and hence he will be able
to sell his piece of vineyard for
1
p
0
p
1
. In total, investing $1 in the vine-
yard will have earned him exactly
q
p
0
p
1
p
0
=
qp
1
p
0
, and we suppose this
amount could have been forecast by the investor with certainty. So there
are two investment possibilities, each one earning for the investor either
1 i (if he chose the nancial asset) or
qp
1
p
0
(if he went for the vineyard)
with certainty. The annual rate of return is
1i1
1
= i in the case of the
nancial asset and,
q p
1
p
0
1
1
=
qp
1
p
0
p
0
=
qhp
p
0
in the case of the vineyard,
where hp Ip
1
p
0
.
We will now show why both assets should earn the same return, and
how markets are going to attain this result through price adjustments.
Suppose rst that the vineyard earns more than the nancial asset (either
in dollar terms or in terms of annual return). Then immediately pure
arbitrage would be in order. Investors would increase their demand for
vineyards; on the nancial market, an excess demand would appear, due
to both a decrease in supply of loanable funds (which prefer to look for a
higher reward in vineyards) and an increase in demand for funds from
pure arbitrageurs who, at no cost, would borrow on the nancial market
in order to invest in vineyards and earn a higher return there. The con-
sequences would be to increase the rate of interest, on the one hand, and
to drive up the price of vineyards, on the other; such an increase in the
price p
0
of the vineyards would decrease the rate of return on the vine-
yard. (The denominator of this rate of return
qhp
p
0
=
qp
1
p
0
1 increases,
and thus the rate of return decreases.) We can see that both rates of return
will converge toward the same value, assuming there are no transaction
costs, because it will be in the interest of investors to enter a market with
the higher return and leave the market with the lower return; this process
will continue until both returns are equal. Hence we have the equality
i =
q
p
0
hp
p
0
(16)
which is called the equation of arbitrage on interest rates, or the Fisher
equation in honor of Irving Fisher, who published it at the end of the
46 Chapter 2
nineteenth century. It may be interesting to note that about two centuries
ago Turgot de l'Aulne came up with this relationship, showing how the
equilibrium prices of capital assets, namely poor lands and fertile lands,
were determined from it (as we mentioned in our introduction).
The current price of the asset p
0
makes this equality hold. In the case we
just considered, we supposed the left-hand side of (16) to be lower than
the right-hand side; this led to an increase in the price of the vineyard, a
decrease in the return on the vineyard, and an increase in the interest rate.
Should the contrary situation have prevailed initially, and the return on
the vineyard have been lower than the interest rate, there would have been
a net supply of vineyards and a net supply of loanable funds on the
nancial market. This would have led to a drop in the price of vineyards,
an increase in their rates of return, and a decrease in the interest rate until
equilibrium was restored.
Until now we have followed our initial hypothesis that both kinds
of assets were riskless, or equivalent from the point of view of our con-
clusions, which rendered the investor neutral to risk. The investment was
riskless if the investor knew with certainty the exact amounts of his rental
rate q and the future price p
1
of his investment (perhaps because he had
been paid his rent in advance and sold his investment in a forward trans-
action). If the investor is neutral to risk, that means he makes a forecast of
q and p
1
(which are random variables) and considers the resulting
expected value E(q p
1
) as if it were a certainty. In either case, the in-
vestor does not attach any weight to the riskiness of his investment. In
reality, we know that certainty can be postulated only if we assume the
existence of forward or futures markets, and very seldom can an invest-
ment be considered as completely riskless; only a handful of individuals
would be truly risk-neutral. Therefore, the Turgot-Fisher equation of
arbitrage i =
qhp
p
0
holds (and permits us to determine the price of capital
assets) only under very special circumstances.
Since investors attach weight to riskiness, they will require a risk pre-
mium over the riskless rate to induce them to hold the risky asset at price
p
0
. In other words, the (expected) rate of return
E(qhp)
p
0
will have to be
higher than i, by a margin called the risk premium, for price p
0
to be an
equilibrium price. Therefore, the current price of the asset p
0
will have to
be lower than the level considered earlier for the following equality to be
valid. Call the risk premium; we will then have:
i =
E(q p
1
)
p
0
1 (17)
and therefore p
0
has to be below its initial value for this equation to be
valid. Figure 2.4 illustrates this point. If p
0
designates the value of p
0
,
which solves i =
E(qp
1
)
p
0
1, then p
/
0
(p
/
0
` p
0
) solves i =
E(qp
1
)
p
/
0
1.
We can of course calculate readily p
/
0
as
E(qp
1
)
1i
and see that the new
price of the asset, taking into account a risk premium, is an increasing
function of the expected rental and the future price, while it is a decreas-
ing function of the interest rate and the risk premium.
The all-important question remains regarding what might be the order
of magnitude for the risk premium . An entire branch of economic and
nance research during these last three decades was devoted to the quest
for an answer to this far from trivial question. In 1990, the Nobel Prize in
economics was awarded to William Sharpe, who in 1964, with his capital
asset pricing model, published the rst attempt to capture this elusive but
essential concept.3
In his path-breaking contribution, Sharpe showed two things. First, the
correct measure of riskiness of an investment was its , that is, the co-
variance of the asset's return with the market return R
M
divided by the
variance of the market return. This is the slope of the regression line
between the market return (the abscissa) and the asset return (the ordi-
nate). Second, the risk premium is the product of and the dierence
between the expected market return E(R
M
) and the risk-free rate i.
According to the capital asset pricing model, is therefore given by
= [E(R
M
) i[
When dealing with risky bonds, we must tackle the dicult issue of the
risk premium attributed by the market to that kind of asset if we want to
know whether the price of the risky bond observed on the market is an
equilibrium price or not. In this text, we deal mainly with bonds bearing
no default risk, so we can safely return to our riskless model for the time
3 See William F. Sharpe, ``Capital Asset Prices: A Theory of Market Equilibrium under
Conditions of Risk,'' Journal of Finance, 19 (September 1964): 425442, or any investment
text (for instance: William F. Sharpe, Gordon J. Alexander, Jerey V. Bailey, Investments,
6th ed., Englewood Clis, N.J.: Prentice Hall International, 1999).
48 Chapter 2
being. However, we will not dodge the problem of the riskiness posed by
possible changes in interest rates.
Coming back, therefore, to our bond market, we realize that the vine-
yard can simply be replaced by a coupon-bearing bond, with the income
stream q generated by the vineyard (the coupon), while the price of the
vineyard p is changed to the price of the bond B. So we must have
i =
c
B

hB
B
(18)
This relationship shows that the return on the bond must be equal to the
rate of interest. In this relationship, caB will always be called the direct
yield of the bond (not to be confused with the coupon rate caB
T
we
introduced earlier). The coupon rate caB
T
is the coupon divided by the
par value (7% in our previous example). The direct yield is the coupon
divided by the current price of the bond B. If this relationship is true, the
price of the bond will see its price change by the amount hBaB = i caB,
which, in our case, is
hB
B
= i
c
B
= 5%
70
1154X44
= 5% 6X06% = 1X06%
E(R) = 1
i +
i
0
E(R)
1
p
0
p
0
p
0
E(q + p
1
)
p
0
E(q + p
1
)
Figure 2.4
Determination of the price of the asset p
//
0
, corresponding to the arbitrage equation of interest
with a risk premium; p
//
0
, is a decreasing function of p.
Let us verify that this is indeed true. When there are 10 years to
maturity, the bond is worth B
0
= 1154X44; one year later, when maturity
is reduced to 9 years, the new price, which can be handily calculated
from formula (13), is 1142.16. Hence the relative price change is
hB
B
=
1142X161154X44
1154X44
= 1X06%, as expected from the equation of arbitrage on
interest.
Before proving rigorously that relationship (16) is true, and that our
numerical results are not just the fruit of some happy coincidence, let us
notice that the direct yield is 6.06%, which is above the rate of interest
(5%). (This is precisely the reason why the owner of the bond will make a
capital loss after one period; there would be a ``free lunch'' otherwise.)
Notice a general property is that the direct yield caB (not only the coupon
rate caB
T
) will always be above the interest rate i if the bond is above par
(equivalently, if the coupon rate is above the rate of interest); this prop-
erty is not quite obvious, and it needs to be demonstrated. Once more,
formula (15) will prove to be very useful. From (15), we know that
B
B
T
is
just a weighted average between
caB
T
i
and 1. So we can always write, when
caB
T
i
is above 1,
caB
T
i
b
B
B
T
b 1
From the rst inequality above, we can write
c
i
b B, and hence
c
B
b i. The
converse, of course, would be true if the coupon rate was below the
interest rate; with caB
T
` i we would have
c
B
` i.
Now, to prove relationship (18), we just have to calculate
hB
B
. Suppose
that B
t
represents the value of the bond at time t with maturity T, and
that B
t1
is the value of the bond at time t 1 when maturity has been
reduced to T 1. All we have to do is calculate
hB
B
=
B
t1
B
t
B
t
. B
t1
is then
the value of the same bond maturing at time T, in T 1 years. Applying
the closed-form expression (13) from section 2.5, we then have
B
t
=
c
i
[1 (1 i)
T
[ B
T
(1 i)
T
=
c
i
(1 i)
T
B
T

c
i
_ _
(19)
B
t1
=
c
i
(1 i)
(T1)
B
T

c
i
_ _
(20)
50 Chapter 2
The change in value of the bond when maturity decreases from T to T 1
is
B
t1
B
t
= B
T

c
i
_ _
[(1 i)
T1
(1 i)
T
[
= B
T

c
i
_ _
[i(1 i)
T
[ (21)
Let us determine the value of this last expression. From the former
denition of B
t
[rst part of (19)], we can write
B
t
=
c
i

c
i
(1 i)
T
B
T
(1 i)
T
iB
t
= c c(1 i)
T
i(1 i)
T
B
T
= c i(1 i)
T
B
T

c
i
_ _
and also
B
T

c
i
_ _
[i(1 i)
T
[ = iB
t
c (22)
Consequently, we get
B
t1
B
t
= iB
t
c (23)
and, nally,
hB
B
=
B
t1
B
t
B
t
= i
c
B
t
(24)
Hence (18) is true.
Of course, as the reader who is already familiar with the concept of
continuous interest rates may well surmise, this relationship holds also
when the bond provides a continuous coupon. It would be written
i(t) =
c(t)
B(t)

1
B(t)
dB(t)
dt
(25)
or, equivalently,
1
B(t)
dB
dt
= i
c(t)
B(t)
We may even add that the demonstration in continuous time is much
easier and much more direct than the demonstration in discrete time. We
will show this in chapter 11, where we deal with continuous spot and
forward rates of return.
2.7 A global picture of a bond's value
We now have a rm grasp of two fundamental properties of bonds. First,
there is a negative relationship between the rate of interest and the price of
the bond; this was pictured in gure 2.3. Second, even in the presence of
uctuations in the interest rate, the price of the bond will ultimately move
toward par. This convergence of the price toward par was pictured in
gure 2.2. We may now want to have a complete picture of these two
phenomena in order to describe how, in the course of its life, the price of
the bond may jump up and down, above and below its par value, but
nevertheless will ultimately end up at its par value. For that purpose, just
consider that either formula (11) or (13) for the bond value is a function
of the two variables i and T, denoted B(iY T), and hence represented by a
surface in a 3-D space where i and T are on the horizontal plane, while
the value of the bond B(iY T) is on the vertical axis (gure 2.5). We have
chosen for that purpose a long bond (20 years) and a coupon of $90, the
par value being 1000.
At any point in time, all the curves representing the value of the bond
as a function of the rate of interest are sections of this surface by planes
parallel to plane (BY i). To illustrate, when the remaining maturity is 20
years, the bond value is 1000, if the rate of interest is 9%. (The bond is
then at par.) But should the rate of interest fall as low as 2%, the price of
the bond would climb as high as 2145. (See the lower half of gure 2.5.)
For a 4% rate of interest, the price of the bond would be 1680. The curve
connecting 2145, 1680, and 1000 is the section of B(iY T) for T = 20; it
represents the value of the bond for all values of the rate of interest be-
tween 2% and 10.5%, when the maturity is 20 years. For shorter matur-
ities, the relevant parallel sections are drawn on the same diagram; they
depict the value of the bond for the same range of interest rates. They are
similar to the curves that were represented in gure 2.3. The reader can
visually verify the convexity of these curves, which become atter and
atter as time passes; ultimately, when the bond is at maturity (when
T = 0) the curve B(i) degenerates to the horizontal line, the height of
52 Chapter 2
i
T
10
20 0%
12%
0
B
9%
6%
3%
B
T
= 1000
B = B
T
+ cT
2800
i
T
10
20 2%
10, 5%
0
B
B
T
= 1000
2145
9%
4%
Figure 2.5
The price of a bond as a function of its maturity and the rate of interest.
which is the par value. The economic interpretation of this horizontal line
is, of course, that whatever may be the interest rate prevailing at the time
the bond reaches maturity, its value will be at par.
The reader can see from the same diagram the evolution of the bond for
any given rate of interest when maturity decreases from T to zero. The
types of curves that were pictured in gure 2.2 are represented for this
bond (9% coupon rate; maturity: 20 years) by sections of the surface in
the direction of the time axis. They are, of course, above the horizontal
plane at height $1000 when the rate of interest is below the coupon rate of
9%, and they lie under it for interest rates above 9%. They all converge
toward the par value of $1000.
We can now nally understand the apparent erratic evolution of the
price of a bond through time. If a bond jumps up and down around its
par value, ending nally at its par value, it is simply because the interest
rate has an erratic behavior. Suppose that the evolution of the rate of
interest is represented by a curve in the horizontal iY T plane, its behavior
being such that it is sometimes above and sometimes below 9%. Also
suppose that its uctuations are more or less steady through time. (In
terms of statistics, we would say that its historical variance is constant.)
Fluctuations of the price of the bond can be understood by watching the
corresponding curve on the surface B(iY T). It can be seen that the uctu-
ations of the price of the bond, which are due solely to those of the rate of
interest, will dampen as maturity approaches, because we are more and
more in the area where the surface gets atter. Although the displace-
ments on the i axis keep more or less the same amplitude, the variations in
the bond's price will become smaller and smaller. Thus it becomes clear
that the historical variance of the price of a bond is not constant, but gets
smaller, tending toward zero when maturity is about to be reached. This
fact will be important if we want to evaluate a bond with a call option
attached to it, which is the case if the issuer of a bond has the right to buy
it back from its owner if the price of the bond increases beyond a certain
xed level (or, equivalently, if rates of interest drop suciently).
2.8 How can we really price bonds?
A rst review of this chapter may leave the reader with the impression
bonds are simply priced as portfolios of zero-coupon bonds, each being
54 Chapter 2
discounted by a rate of interest that acts erratically and unpredictably.
First, we considered a ``given'' term structure, and then we simplied this
to a horizontal (constant) term structure. In reality, however, discount
factors and the term structure are themselves computed from the
universe of bond prices in a number of ways that will be explained later
in the book (in chapter 9). We will also address some of the diculties
researchers have met in doing so.
This leads us to another problem. Imagine that a sophisticated econo-
metric method has been devised to obtain from the universe of bond
prices a complete series of spot rates (the entire term structure). We
should then raise the following issue: if bond prices are equilibrium prices,
then the corresponding term structure can certainly be considered one. In
theory, this information could then be used to determine if any given
triple-A bond is under- or overpriced. But how can we be sure these are
equilibrium values? And even if we knew for sure what the true equilib-
rium values are, we would not yet be safe. As John Maynard Keynes once
pointed out, what we need to know is not what assets' true equilibrium
values are, but what the market thinks the equilibrium values are! What
Keynes said of stocks is of course equally relevant for bonds, and we are
still a long way from making sure bets on the direction that bond prices
or spot rates will take in the future, and even further from guessing then
exactly.
Key concepts
.
The value of a bond is the sum of the discounted cash ows of the bond
.
A bond is above par if its coupon rate is above interest rates, and vice
versa.
.
The value of a default-free bond will always decrease if interest rates
rise, and vice versa.
.
The fundamental determinant of changes in the price of a bond is
movement in interest rates; a secondary cause is shortening maturity
Basic formulas
.
Notation: B = value of bond today; T = maturity; B
T
= par or re-
demption value; c = coupon value; i
t
= spot rate corresponding to term t;
i = interest rate pertaining to all maturities; c
t
= cash ow received by
owner at time t.
.
B(i
t
) =

T
t=1
c
t
(1 i
t
)
t
, t = 1Y F F F Y T (i
t
denes the ``term structure of
interest rates'')
.
B(i)aB
T
=
caB
T
i
1
1
(1i)
T
_ _
1
(1i)
T
(if i
t
is constant for all terms t, that
is, the term structure is at).
.
hB
B(i)
= i
c
B(i)
Questions4
2.1. Consider a bond whose par value is $1000, its maturity 10 years, and
its coupon rate 9%. (Assume that the coupon is paid once a year.) Sup-
pose also that the spot interest rates for each of those 10 years is constant
(the term structure of interest rates is said to be at), and consider that
their level is either 12%, 9%, or 6%.
a. Without doing any calculations, explain whether the bond's value will
be increasing or decreasing when interest rates decrease from 12% to 9%
to 6%. In which of these three cases can the bond's value be predicted
without any calculations?
b. In which case will the bond's value be the farthest from its par value,
and why?
c. Conrm your answers to (a) and (b) by determining the bond's value in
each case.
2.2. Consider two bonds (A and B) with the same maturity and par value.
Bond B, however, has a coupon (or a coupon rate) which is 20% lower
than that of A. Will the price of bond B be
a. lower than A's price by more than 20%?
b. lower than A's price by 20%?
c. lower than A's price by less than 20%?
Explain your answer. Can you gure out a case in which the price of B
would be lower than A's price by exactly 20%?
4 In the questions for this chapter and the next ones, simply consider that coupons are paid
once a year.
56 Chapter 2
2.3. In a riskless world, suppose that the future (or futures or forward)
price of an asset with a one-year maturity contract is $100,000; the (cer-
tain) rental rate of that asset is $5000, while the riskless interest rate is 6%
per year. What is the equilibrium expected rate of return on this asset?
What is the current price of this asset?
2.4. Using the same data as in question 2.3, we now assume the market
applied a 10% risk premium to that kind of investment. What would be
the equilibrium price of the asset in this risky world?
2.5. Suppose that one of the following events for an investment occurs:
a. The expected rental of the asset increases.
b. The expected future price of the asset increases.
c. The current price of the asset increases.
If the risk premium of the asset does not change, what are the con-
sequences of these events on the expected rate of return of the investment?
3
THE VARIOUS CONCEPTS OF RATES OF
RETURN ON BONDS: YIELD TO MATURITY
AND HORIZON RATE OF RETURN
Shylock. Is it so nominated in the bond?
. . .
I cannot nd it; 'tis not in the bond!
Chapter outline
3.1 Yield to maturity 60
3.2 Yield to maturity for bonds paying semi-annual coupons 61
3.3 The drawbacks of the yield to maturity as a measure of the bond's
future performance 62
3.4 The horizon rate of return 62
Appendix 3.A.1 Yield to maturity for semi-annual coupons, and any
remaining maturity (not necessarily an integer number of periods) 66
Overview
There are two fundamental rates of return that can be dened for a bond:
the yield to maturity and the horizon rate of return. The yield to maturity
of a bond is entirely akin to the internal rate of return of an investment
project: it is the discount rate that makes equal the cost of buying the
bond at its market price and the present value of the future cash ows
generated by the bond. Equivalently, it can be viewed as the rate of return
an investor would make on an investment equal to the cost of the bond if
all cash ows could be reinvested at that rate.
The concept of future value is central to that of horizon rate of return.
If all the proceeds of a bond are reinvested at certain rates until a given
horizon, then a future value of investment can be calculated with respect
to that horizon. It is made up of two parts: rst, the future value of the
cash ows reinvested at various rates (which have to be determined); sec-
ond, the remaining value of the bond at that horizon, which consists of the
remaining coupons and par value expressed in terms of the horizon year
chosen. From that future value at a given horizon, it is possible to dene
the horizon rate of return as the yearly rate of return, which would trans-
form an investment equal to the cost of the bond into that future value.
There are advantages and some inconveniences in using each kind of
rate of return. The advantage of using the internal rate of return is that it
is very easy to calculate (all software programs perform this calculation
handily); the problem with that concept is that we have to suppose,
implicitly, that all proceeds of the bond will be reinvested at that same
ratewhich, of course, is not always the case. This is why the concept of
horizon rate of return was introduced: to take into account the fact that the
entire structure of interest rates is permanently moving. The increase in
realism has its cost; we have to be able to forecast what the future spot rates
will be for all the cash ows reinvested until the horizon year, and to fore-
cast as well the structure of interest rates prevailing at the horizon year for
the remaining cash ows to be received at that timeno light task indeed.
3.1 Yield to maturity
The yield to maturity of an investment that buys and holds a bond until its
maturity is entirely akin to the concept of the internal rate of return of a
physical investment. It shares its simplicity but also, unfortunately, its
drawbacks.
Suppose that an investmentbe it physical or nancialis charac-
terized by a series of cash ows that are supposed to be known with cer-
tainty; this is precisely the case of a defaultless bond. Suppose also that
we know the price today to be B
0
, and that the bond will be kept to its
maturity. The question then arises regarding the rate of return this bond
will earn for its owner. Traditionally, one way to answer this question is
to dene the bond's yield to maturity in the following way: it is the rate of
interest that will make the present value of the bond's cash ows exactly
equal to the price of the bond. Thus the yield to maturity, denoted here i
+
,
will be the rate of interest such that
B
0
=
c
1 i
+

c
(1 i
+
)
2

c B
T
(1 i
+
)
T
(1)
Written in more compact form, i
+
is such that
B
0
=
T
t=1
c
t
(1 i
+
)
t
(1
/
)
where c
t
(t = 1Y F F F Y T) are the cash ows of the bond.
How do we calculate this value i
+
? Analytically, in the general case, it is
not possible to calculate a closed-form expression for i
+
depending on B
0
,
60 Chapter 3
c, and B
T
. Indeed, should we try to do so, we would soon be on our way
to calculating the roots of a polynomial of order Tand for T b 4, no
closed form of these roots exists. Therefore, only a trial-and-error process
can be used, such as those developed today in pocket calculators or in any
nancial software; for one initial value of i, the computer calculates the
present value of the sum of the cash ows. If it happens to be above B
0
,
the computer tries a higher value for i; should the present value be lower
than B
0
this time, the computer will choose a somewhat lower value for i.
This process continues very quickly until a value of i gives B
0
with a suf-
ciently high accuracy; that will be i
+
, the yield to maturity of the bond.
Dening the yield to maturity in such a way, as that value i
+
that
equates both sides of equation (1), has the drawback of concealing the
true signicance of the return that is supposed to accrue to the bond's
owner. Note that the yield to maturity i
+
is also the return to the investor,
which makes it equivalent to buying the bond, receiving and reinvesting
all coupons at rate i
+
, and investing a lump-sum B
0
at rate i
+
for T years.
This can be seen by multiplying both sides of (1) by (1 i
+
)
T
(we just
transform present values on both sides into future values at horizon T).
We get
B
0
(1 i
+
)
T
= c(1 i
+
)
T1
c(1 i
+
)
T2
c B
T
(2)
which translates as
Future value of an investment B
0
= future value of all cash ows reinvested at rate i
+
3.2 Yield to maturity for bonds paying semi-annual coupons
As we mentioned in chapter 1, the price the investor in a Treasury note or
bond receives is not just the ask price, but the ask price plus any accrued
interest (AI), if applicable; that is if at the time of the transaction some
time has elapsed since the last coupon payment or since the bond's issu-
ance (as in most cases, of course), when no coupon has yet been made.
We call ``total price'' (TP) the sum of the ask price (AP) and accrued
interest (AI). So we have TP = AP AI. Let n be the number of periods
(for instance, six-month periods) remaining in the bond's lifeor until
the rst call date established by the bond's issuer. Let f be the fraction of
the period remaining until the next coupon. For a bond paying a semi-
annual coupon, the yield to maturity of that bondits internal rate to its
61 The Various Concepts of Rates of Return on Bonds
potential investoris the rate y, which equates its cost to the present
value of all future discounted cash ows. (In economic parlance, the
interest rate makes its net present value equal to zero.) The yield to ma-
turity is thus y, such that it solves the equation:
TP =
f (ca2)
(1 ya2)
f

n
t=1
ca2
(1 ya2)
tf

par
(1 ya2)
nf
As before, only trial-and-error or numerical methods can solve this
equation. Numerous software programs and even pocket calculators
make such calculations quite easy; as inputs, one simply needs the date of
purchase, the maturity (or rst call date), and the yearly coupon to obtain
the yield to maturity.
3.3 The drawbacks of the yield to maturity as a measure of the bond's future
performance
The central problem with the concept of yield to maturity is simply that it
is a return for the owner of the bond if we assume that all cash ows will
be reinvested at the same rate i
+
, which is hardly realistic, precisely be-
cause we know that interest rates do move around. After all, when we
dened the yearly return on a bond, this was equal to
c
B

hB
B
.
c
B
. The direct
yield was known with certainty, but hB, the future price change in the
bond, was not known, precisely because the future rates of interest, one
year from now, were unknown. It is therefore no surprise that the yield to
maturity of a bond, which usually involves an investment period longer
than one year, does not accurately predict what the yield for the investor
will be. A better measure is needed here, and this is precisely why the
concept of horizon rate of return (also called the holding period rate of
return) has been introduced, and is fortunately more and more widespread.
3.4 The horizon rate of return
The purpose of the horizon rate of return is to measure the future per-
formance of the investor if he keeps his bond for a number of years (called
the horizon). Dened in the most general way, the horizon rate of return,
denoted as r
H
, is the return that transforms an investment bought today
at price B
0
by its owner into a future value at horizon H, F
H
. Thus r
H
is
such that
62 Chapter 3
B
0
(1 r
H
)
H
= F
H
(3)
and r
H
is equal to
r
H
=
F
H
B
0
_ _
1aH
1 (4)
As the reader may well surmise, the diculty here lies in calculating the
future value F
H
of the investment when this investment is a coupon-
bearing bond. Indeed, all coupons to be paid are supposed to be regularly
reinvested until horizon H (which may well be, of course, shorter than
maturity T). Therefore, we have to be in a position to evaluate the future
rates of interest at which it will be possible to reinvest these coupons. This
implies a forecast of future rates of interest, which is no easy task. For the
time being, to keep matters clear, consider the situation of an investor
who has just bought our bond valued at B(i
0
) when the interest rate
common to all terms was equal to i
0
. Suppose the following day the in-
terest rate moves up to i (i Hi
0
); suppose also that an investor holds onto
his investment for H years. What will be his H-year horizon return if we
suppose that interest rates will remain xed for that time span? The new
value of the bond, when i
0
moves to i, will be B(i), and its future value
will be F
H
= B(i)(1 i)
H
. Hence, applying our formula (4), we get
r
H
=
B(i)
B
0
_ _
1aH
(1 i) 1 (5)
As an example, consider the situation of our investor who has bought
the bond with a remaining term to maturity of 10 years and a coupon rate
of 7%, when interest rates were at 5%. He has paid B
0
= B(i
0
) = $1154X44.
Suppose now that the next day interest rates move up to 6%. His bond
will drop in value to $1073.60. If he holds his bond for 5 years, and if
interest rates stay at 6% for that span of years, his horizon return will be
r
H
=
1073X60
1154X44
_ _
1a5
(1X06) 1 = 4X47%
We observe immediately that, although the rate of interest at which he
can reinvest his coupons (6%) is higher than initially, his overall yearly
performance will be smaller than 5%. This apparently strange phenome-
non can be explained by the capital loss our investor has incurred when
the interest rate moved up from 5% to 6%, and by the fact that the re-
investment of coupons at a higher rate (6%) than the previous one (5%)
has not been sucient to compensate for a capital loss that still hurts
after 5 years. When such an increase in interest rates happens, two con-
icting phenomena result: rst, there is a capital loss and, second, there
is some gain from the reinvestment of coupons at a higher rate. Which
of these two phenomena will prevail? In other words, will the horizon rate
of return be higher or lower than the initial yield at which the bond was
bought? From what we have said already (and for the time being!),
nobody can tell, unless we make the direct calculations as indicated by
formula (5). But one thing is sure: the longer the horizon and the longer
the replacement of coupons at a higher rate, the greater the chances that
the investor will outperform the initial yield of 5%.
This will be true for two reasons: First, the reinvestment of coupons for
a long period of time will tend to boost the horizon rate of return; second,
the longer the horizon, the closer to its par value the bond will be, thereby
osetting the initial loss in value due to the rise in the interest rate. Of
course, the reverse would be true if interest rates fall. Suppose that interest
rates fall from 5% to 4% immediately after the purchase of the bond. The
investor will realize an important capital gain, which will erode itself as
time passes. On the other hand, the investor with a short horizon will not
have to care too much about the reinvestment of his coupons at a lower
rate (4%) than the original rate (5%). Thus, in the event of a decrease in
interest rates, the shorter the horizon, the higher the performance (there
are important capital gains and little to lose in coupon reinvestment); and
the longer the horizon, the smaller the performance. (The capital gain will
dampen out through time, and the loss from smaller returns on coupon
reinvestment will pinch more.)
Before getting a little deeper into this subject, let us show, as an exam-
ple, what the horizon return would be in the three following scenarios: no
change in the rate of interest (5%), an increase of the rate of interest to
6%, and a decrease to 4%. The bond is the one we have already consid-
ered in our example: coupon 7%; maturity: 10 years; par value: 1000.
Our intuitive grasp of the properties of the horizon rate of return are
conrmed when reading table 3.1. The longer the horizon, the higher the
horizon return if interest rates move up immediately after the bond's
purchase, and the lower the horizon return if interest rates have gone
64 Chapter 3
down. Of course, the reader will want to have sound proof of this propo-
sition by considering formula (5). If interest rates go up, B(i)aB(i
0
) is
smaller than one, and the Hth root of a number smaller than one is also a
number smaller than one, and it is an increasing function of H. Con-
versely, if interest rates go down, B(i)aB(i
0
) is larger than one, and its Hth
root is a decreasing function of H. Therefore, the horizon rate of return,
dened by (5), is an increasing function of the horizon H if i b i
0
, and a
decreasing function of H if i ` i
0
(it is equal to i
0
and independent from
H if i stays at i
0
).
We now need to notice another remarkable property of these horizon
rates of return. We know that in each interest scenario, the horizon return
may be smaller than 5% or higher than that value, depending on the
horizon itself. It turns out that in both scenarios there exists a horizon for
which the horizon return will be extremely close to 5%. This horizon
seems to be between seven and eight years, and, at rst inspection of the
table, perhaps closer to eight years than seven. Indeed, some calculations
show that if the horizon is 7.7 years, then the horizon rate of return will be
slightly above 5% (5.006%) whether the interest rate falls to 4% or
increases to 6%. Immediately, the question arises: Is this a coincidence
stemming from the special features of our bond, or would it be possible
for any bond to have a horizon such that the horizon return is equal to the
original rate of return whatever happens to the rates of interest? This
question is important. We will show later on that this unique horizon does
Table 3.1
Horizon return (in percentage per year) on the investment in a 7% coupon, 10-year maturity
bond bought 1154.44 when interest rates were at 5%, should interest rates move immediately
either to 6% or to 4%
Interest rates
Horizon (years) 4% 5% 6%
1 12.01 5 1.42
2 7.93 5 2.22
3 6.60 5 3.47
4 5.95 5 4.09
5 5.55 5 4.47
6 5.29 5 4.73
7 5.11 5 4.92
7.7 5.006 5 5.006
8 4.97 5 5.04
9 4.86 5 5.15
10 4.77 5 5.23
in fact exist for any bond or bond portfolio; its name is the duration of the
bond. This concept and its properties are so important that we will now
devote a special chapter to it.
Appendix 3.A.1 Yield to maturity for semi-annual coupons, and any remaining
maturity (not necessarily an integer number of periods)
The calculation of the yield to maturity of a bond paying semi-annual
coupons, and whose next coupon will be paid in less than half a year, can
be made in a number of alternative ways with the same result. We nd the
simplest way to be the following.
Let `òne period'' denote six months, and assume that one coupon is
paid at the end of each period.
Let 0 designate the point in time at which the last coupon was paid or
the date of the bond's issuance if no coupon has yet been paid. We will
call this point in time `ìnitial time.''
Let 0 be the fraction of a period elapsed since the last coupon payment
or since the date of the bond's issuance if no coupon has yet been paid.
Thus, if a coupon is paid twice a year, we always have 0 0 ` 1 (see
gure 3.1).
Let n be the number of coupons remaining to be paid. From our de-
nitions and gure 3.1, this implies that the next coupon will be paid in
1 0 half-years. If i is the annual rate of interest used to discount all cash
ows (coupons and principal) to be paid, the present value of those cash
ows as viewed from the initial time is
Value of bond at time 0 =
n
j=1
c
j
(1 ia2)
j
(6)
where c
j
designates the jth cash ow and j designates the periods.
Now this value at point 0 should be translated in value at point 0. We
do this simply by multiplying the above expression by (1
i
2
)
0
, where 0 is
the fraction of the six-month period elapsed since initial time 0. (If 31
days have elapsed in a 182-day six-month period, then 0 is simply
31
182
= 0X1703.) We have then
Value of bond at time 0 = 1
i
2
_ _
0
n
j=1
c
j
(1 ia2)
j
(7)
66 Chapter 3
Now remember how an internal rate of return is calculated: it is the rate
of interest that equates the cost of an investment to its discounted net cash
ows; equivalently, it is the rate of interest that makes the net present
value of the investment equal to zero. Here the cost of the investment is
the cost of the bond (the amount paid by the buyer), and it is equal to the
quoted price of the bond plus the accrued interest on the bond. This
accrued interest belongs to the initial owner of the bond at the time he
sells it; so the buyer redeems it for 0c. Thus, the cost of acquiring the
bond is
Cost to the buyer = ask price 0c
The yield to maturity is thus the rate of interest i such that the following
equality holds:
Cost to the buyer = present value of cash flows at time 0
= ask price 0c = (1 ia2)
0
n
j=1
c
j
(1 ia2)
j
= (1 ia2)
0
2c
i
1
1
(1 ia2)
n
_ _
B
T
(1 ia2)
n
_ _
As before, this equation does not generally have a closed-form solution,
and only trial-and-error solutions, usually embedded in pocket calculators
or spreadsheets, are available. This yield to maturity is called `àsk yield''
Figure 3.1
Correspondence between time and the number of cash ows remaining to be paid for a bond
paying coupons twice a year.
because it it based on the price that is asked to the potential customer (the
price he will have to pay, in addition to the accrued interest 0c).
Note that if the value of the bond at time 0 had been calculated by
multiplying
n
j=1
c
j
(1 ia2)
j
by 1 0ia2 instead of (1 ia2)
0
(that is,
by using a simple-compounding interest formula instead of a com-
pounded one), nothing much would have changed because 1 (ia2)0 is
just the rst-order Taylor approximation of (1 ia2)
0
at point i = 01;
also ia2 is a small number compared to zero. Consider, for instance, our
former example, where one month has elapsed since the last coupon has
been paid out or since the bond was emitted. In that case, 0 =
31
182
=
0X1703. Suppose also that i = 5%/year. Applying (1 ia2)
0
, we get
1X025
31a182
= 1X0042; on the other hand, 1 (ia2)0 yields 1 (0X025)
31
182
= 1X0042; the rst four digits are caught by the McLaurin approxima-
tion. If we consider the longest period that 0 could ever take, that is 0 =
181a182, we would have (1 ia2)
0
= 1X02486 and 1 (ia2)0 = 1X02486
(even the rst ve decimals would be caught). In the case of the shortest
period possible for 0, 1a182, we would have (1 ia2)
1a182
= 1X00013 and
1 (ia2)(1a182) = 1X00013, no dierence to speak of either.
Key concepts
.
Yield to maturity is the interest rate that makes the price of the bond
equal to the sum of its cash ows, discounted at this single interest rate.
.
Horizon rate of return is the interest rate that transforms the present
cost of an investment into a future value.
Basic formulas
.
Yield to maturity: i
+
such that
B
0
=

T
t=1
c
t
(1 i
+
)
t
where B
0
is the bond's value today and c
t
are the annual cash ows.
1 Indeed, we have (1 ia2)
0
e1 (ia2)0 as a rst-order MacLaurin approximation (the
Taylor approximation at point ia2 = 0).
68 Chapter 3
.
Horizon rate of return: with F
H
Ifuture value of the investment, r
H
is
such that
B
0
(1 r
H
)
H
= F
H
r
H
= (
F
H
B
0
)
1aH
1
Questions
3.1. Why did we say, in section 3.1, that in order to calculate analytically
the yield to maturity of a coupon-bearing bond with maturity T, we
would have to solve a polynomial equation of order T?
3.2. What is the horizon rate of return on a bond if the interest rate does
not move and stays xed at i
0
?
3.3. A bond with 9% coupon rate, $1000 par value, and 30-year maturity
was bought when the (at) interest structure was 8%. The next day, the
interest rates jumped to 9% and stayed at that level for 10 years. What are
the horizon returns on that bond for the following horizons H?
a. H = 1 year
b. H = 4 years
c. H = 10 years
3.4. Consider the scenario corresponding to section 3.3. What would be
the innite horizon rate of return on an investment in such a scenario?
Give an answer from analytical considerations, and then give an answer
that does not require any calculations.
3.5. Using a spreadsheet perhaps, conrm the results established in table
3.1. (Suppose as usual that coupons are paid yearly.)
3.6. Referring to section 3.3, can you explain intuitively the existence of a
horizon leading to the same rate of return whenever interest rates either
increase from 5% to 6% or decrease to 4%?
4
DURATION: DEFINITION, MAIN
PROPERTIES, AND USES
Portia. . . . 'tis to peize the time1
Chapter outline
4.1 Denition and calculation 72
4.2 Duration as a center of gravity 75
4.3 Is duration an important concept? 75
4.4 Duration as a measure of a bond's sensitivity to changes in yield to
maturity 76
4.5 The concept of modied duration 78
4.6 Properties of duration 79
4.6.1 Relationship between duration and coupon rate 80
4.6.2 Relationship between duration and yield 81
4.6.3 Relationship between duration and maturity 83
4.7 Duration of a bond portfolio 86
4.8 Measuring the riskiness of foreign currencydenominated bonds 87
Overview
There are two main reasons why duration is an essential concept in bond
analysis and management: it provides a useful measure of the bond's
riskiness; and it is central to the problem of immunization, that is, the
process by which an investor can seek protection against unforeseen
movements in interest rates.
Duration is simply dened as the weighted average of the times of
the bond's cash ow payments; the weights are proportions of the cash
ows in present value. This concept has a nice physical interpretation,
as the center of gravity of the cash ows in present value. This interpre-
tation proves to be very useful when we want to know how duration
1French-speaking readers will have little diculty understanding the verb ``to peize'' used by
Portia, because the Latin pensare gave the French language both penser (to think) and peser
(to weigh). It is now interesting for both French- and English-speaking readers to look up
this verb in The Shorter Oxford English Dictionary (2 Vol., Oxford University Press, 1973, p.
1540). We nd: ``3. To keep or place in equilibrium.'' Placing time in equilibrium is nothing
short of determining duration, as we will see in this chapter. Once again Shakespeare had
said it all.
will be modied if either the interest rate, the coupon rate, the maturity of
the bond, or any combination of those three factors are modied.
The denition of the duration of a bond extends immediately to the
denition of the duration of a bond portfolio: this is simply the weighted
average of all bonds' durations, the weights being the shares of each bond
in the portfolio.
Duration can be shown to be equal to the relative increase, by linear
approximation, in the bond's price multiplied by minus the factor rate of
capitalization (dened as one plus the rate of interest). This gives a fair
approximation of the riskiness of the bond, in the sense that we immedi-
ately know from the bond's duration the order of magnitude by which the
bond's price will decrease (in percentage) when the rate of interest
increases by one percent.
Duration is signicantly dependent on the coupon rate, the rate of
interest, and maturity. An increase in the coupon rate will lead to a
decrease in duration. (Such an increase in the coupon rate will enhance
the importance of early payments; therefore, the weights of the rst terms
of payment will be increased relatively to the last onewhich includes the
principaland consequently, the center of gravity of these weights, which
is equal to duration, will move to the left.) An increase in the rate of
interest will have a similar eect on duration, because the present value of
the payments will be signicantly diminished for late payments, much less
so for early ones; thus the center of gravity of those payments will also
move to the left. The eect of increasing the maturity of a bond on its
duration is more complex, and contrary to what our intuition would
suggest to us, increasing maturity will not necessarily increase duration.
In special cases, for above-par bonds and very long maturities, the con-
verse will be true. We will explain why this is so.
Finally, it should be emphasized that, in the case of foreign currency
denominated bonds, the greater part of the risk lies in modications of the
exchange rate.
4.1 Denition and calculation
.
Denition: the duration of a bond is the weighted average of the times of
payment of all the cash ows generated by the bond, the weights being the
shares of the bond's cash ows in the bond's present value.
72 Chapter 4
Not surprisingly, this denition extends to the duration of a portfolio of
bonds. If we consider a portfolio of bonds with the same yield to maturity,
the duration will be the weighted average of the times of payment of all the
cash ows generated by the portfolio (the weights being the shares of
the cash ows in present value). We will show later that this duration is the
weighted average of the durations of the bonds making up the portfolio.
As seen in chapter 1, there are several ways to calculate the present
value of any cash ow received in the future. One way is to calculate the
present value of the cash ows (the sum of which will equal the bond's
price). If we use the internal rate of return as the discount rate of the cash
ows, the duration will be called Macauley's duration, in honor of Fred-
erick Macauley who developed this concept in 1938.2 The concept of du-
ration was independently rediscovered by Paul Samuelson (1945), John
Hicks (1946), and F. M. Redington (1952); Samuelson and Redington
invented the concept in their analysis of the sensitivity of nancial insti-
tutions' assets and liabilities to changes in interest rates.3 For a historical
presentation of the subject, as well as its development, see Gerald Bierwag
(1987) and G. Bierwag, G. Kaufman and A. Toevs (1983).4 If, on the
other hand, we use spot rates to evaluate the cash ows, we will get an-
other set of weights for our average (although the result will, generally, be
very close to that obtained with the Macauley method). If we use spot
rates, we will get a duration referred to as the Fisher-Weil duration.5 We
will start with the Macauley concept and then consider the Fisher-Weil
approach when dealing, later on, with the immunization process.
2Frederick Macauley, ``Some Theoretical Problems Suggested by the Movements of Interest
Rates, Bond Yields and Stock Prices in the United States since 1865,'' National Bureau of
Economic Research, 1938.
3The reader should refer to P. A. Samuelson, ``The Eect of Interest Rate Increases on the
Banking System,'' The American Economic Review (March 1945): 1627; T. R. Hicks, Value
and Capital (Oxford: Clarendon Press, 1946), and F. M. Redington, ``Review of the Principle
of Life Oce Valuations,'' Journal of the Institute of Actuaries, 18 (1952): 286340.
4G. Bierwag, Duration Analysis: Managing Interest Rate Risk (Cambridge, Mass.: Ballinger
Publishing Company, 1987), and G. Bierwag, G. Kaufman, and A. Toevs, ``Recent Devel-
opments in Bond Portfolio Immunization Strategies,'' in Innovations in Bond Portfolio
Management, ed. G. Kaufman, G. Bierwag, and A. Toevs (Greenwich, Conn.: TAI Press,
1983).
5Lawrence Fisher and Roman Weil, ``Coping with the Risk of Interest-rate Fluctuations:
Return to Bondholders from Naive and Optimal Strategies,'' The Journal of Business, 44,
no. 4 (October 1971).
73 Duration: Denition, Main Properties, and Uses
The Macauley duration is calculated as follows. When i denotes the
yield to maturity of the bond, since it is the weighted average of all times
of payment t = 1Y 2Y F F F Y T, we can write
D = 1
caB
1 i
2
caB
(1 i)
2
T
(c B
T
)aB
(1 i)
T
(1)
or, more compactly,
D =
T
t=1
t
c
t
aB
(1 i)
t
=
1
B
T
t=1
tc
t
(1 i)
t
(2)
where B designates the value of the bond, c
t
is the cash ow of the bond
to be received at time t, and i is the bond's yield to maturity (which we
had previously called i
+
but now simply call i for convenience).
Table 4.1 shows how the duration of the bond can be calculated for our
7% coupon bond, when all the spot interest rates are 5% (alternatively,
when the yield to maturity is 5% and when the price of the bond is
$1154.44). Thus, the bond has a duration of 7.7 years. This means that the
weighted average of the times of payment for this bond is 7.7 years when
the weights are the cash ows in present value.
Table 4.1
Calculation of the duration of a bond with a 7% coupon rate, for a yield to maturity i F5%
(1)
Time of
payment t
(2)
Cash ows in
current value
(3)
Cash ows in
present value
(i = 5%)
(4)
Share of cash
ows in present
value in bond's
price
(5)
Weighted time
of payment
(col. 1 col. 4)
1 70 66.67 0.0577 0.0577
2 70 63.49 0.0550 0.1100
3 70 60.47 0.0524 0.1571
4 70 57.59 0.0499 0.1995
5 70 54.85 0.0475 0.2375
6 70 52.24 0.0452 0.2715
7 70 49.75 0.0431 0.3016
8 70 47.38 0.0410 0.3283
9 70 45.12 0.0391 0.3517
10 1070 656.89 0.5690 5.6901
Total 1700 1154.44 1 7.705 = duration
74 Chapter 4
4.2 Duration as a center of gravity
Duration has a neat physical interpretation. As a weighted average, it is
none other than a center of gravity; it will be the center for gravity for the
cash ows in present value. If these cash ows, represented as the hatched
rectangles in gure 2.1 (chapter 2), are considered as weights, then the
duration of our bond is the center of gravity of the (weightless) ruler that
would support them. Figure 4.1 presents this physical analogy, which will
be very useful because we will be able to derive from it some important
properties of duration.
4.3 Is duration an important concept?
When asked what they know about duration, most people will respond
that it constitutes a measure of the riskiness of the bond with respect to
changes in the yield to maturity of the bond. This does not, however,
constitute the main interest of the concept. Duration can, indeed, be used
0
700
600
500
400
300
200
100
1 2 3 4 5
Years of payment
Duration
7.705 years
6 7 8 9 10
C
a
s
h
f
l
o
w
s

i
n

p
r
e
s
e
n
t

v
a
l
u
e
Figure 4.1
Duration of a bond as the center of gravity of its cash ows in present value (coupon: 7%;
interest rate: 5%).
as a proxy of the bond price sensitivity to changes in the yield. But, sup-
posing that knowing this sensitivity with regard to this statistic (the de-
ciencies of which have already been shown) is important, computers or
even hand calculators can almost immediately give the exact sensitivity of
the price to changes in yield.
The true reason why such a concept is so important is that, originally
devised by actuaries, it is today at the heart of the immunization process, a
strategy designed to protect the investor against changes in interest rates.
We will devote an entire chapter to this strategy (chapter 6).
Duration and the immunization process are also highly relevant to
banks, pension funds, and other nancial institutions. In chapter 7, we
will give some important applications of these concepts, and we will
address the problem of matching future assets and liabilities (dened for
the rst time by F.M. Redington in 1952). We will then turn to hedging a
nancial position with futures. Each time duration will play a central role.
Finally, as we will show in the next chapter, duration can help clarify a
rather arcane concept for the nonspecialist: the bias introduced in the
T-bond futures conversion factor. However, for the time being, we will
merely analyze how and why duration is a measure of bond price sensi-
tivity to interest rates. We will then proceed to analyze the properties of
duration because they and their economic interpretations will prove to be
very interesting, often surprising, and highly useful.
4.4 Duration as a measure of a bond's sensitivity to changes in yield to maturity
Let us now show that duration is a measure of the riskiness of the bond.
More precisely, we will show that duration (denoted D) is, in linear ap-
proximation, the relative rate of decrease in the value of the bond when
the yield to maturity i increases by 1%, multiplied by 1 i. We have
D =
1 i
B
dB
di
(3)
The simplest way to show this is to write the value of the bond as
B =
T
t=1
c
t
(1 i)
t
(4)
76 Chapter 4
and take the derivative of this expression with respect to i. (Remember
that i can be considered either as the yield to maturity of the bond, or a
spot rate common to all maturities t = 1Y F F F Y T.) We will have
dB
di
=
T
t=1
(t)c
t
(1 i)
t1
=
1
1 i
T
t=1
tc
t
(1 i)
t
(5)
Dividing (5) by B, we get
1
B
dB
di
=
1
1 i
T
t=1
tc
t
(1 i)
t
aB (6)
We recognize that the right-hand side of (6) contains the expression of
duration given by (2). We can now write
1
B
dB
di
=
D
1 i
(7)
which shows that the relative increase in the value of the bond is minus its
duration divided by 1 plus the yield to maturity. Equivalently, we have
D =
1 i
B
dB
di
(8)
which is what we wanted to show. Both expressions, (7) and (8), will be of
utmost importance.
Let us consider some examples. Suppose a bond is at par (B = B
T
=
100); its coupon is 9%, and therefore the yield to maturity is 9%. (When
you consider equation (13) in chapter 1 remember that i = caB
T
is the
only way that B = B
T
, or BaB
T
= 1.) Its duration can be calculated
as 6.995 years. Suppose that the yield increases by 1%; we then have, in
linear approximation,
hB
B
e
dB
B
=
D
1 i
di =
6X99
1X09
1% = 6X4%
(Notice that the result is indeed expressed as a percentage, because the
time units of duration Dexpressed in yearscancel out those of di,
which are percent per year, and 1 i is a pure number.)
This result is highly useful, because duration gives us a very quick way
to approximate the change in the value of the bond if interest rates move
by 1%: it suces to divide duration by 1 i to get the approximate
decrease of the bond's price when i moves up by 1%.
How good is the approximation? It is a rather satisfactory one, because
the relationship between B and i is nearly a straight line, and therefore the
derivative of B with respect to i is very close to its exact rate of increase.
Indeed, should we calculate the latter, we would get in this case
hB
B
=
B(i
1
) B(i
0
)
B(i
0
)
=
B(10%) B(9%)
B(9%)
=
93X855 100
100
= 6X145%
as opposed to 6.4%. Notice, however, that the linear approximation to
the curve B(i) conceals the all-important phenomenon of the convexity of
that curve. Duration analysis tells us that the bond will either increase or
decrease approximately by 6.4% in case of a decrease or a 1% increase in
yield i, respectively. However, in exact value the bond will decrease by less
than 6.4% (in fact, by 6.145%), and it will increase by more than that
amount if i moves down to 8%; the bond's price increase will be
B(8%)B(9%)
B(9%)
= 6X71%. This, of course, is the direct result of convexity of
bonds (a property we have already mentioned in chapter 1), a topic so
important we will devote an entire chapter to it.
4.5 The concept of modied duration
Since duration is so closely linked to the change in price of a bond,
D
1i
has
a special name: modied duration, and we shall denote it by D
m
. Modied
duration is simply duration divided by one plus the yield to maturity. We
have
Modied duration ID
m
=
duration
1 i
and therefore,
modied
hB
B
e
dB
B
= duration di = D
m
di
78 Chapter 4
In our current example, modied duration is 6.4 years. Coming back to
the example we considered earlier (a bond with coupon rate 70, par 1000,
and yield 5%), duration was calculated as 7.7 years; modied duration is
then 7X7a1X05 = 7X33 years. This means that a change in yield from 5% to
either 6% or 4% entails a relative change in the bond's price approxi-
mately equal to 7.33% or 7.33%, respectively.
It is therefore quite natural to consider duration to be a measure of the
interest rate risk of the bond, because if a high duration for a bond implies
that its owner will highly benet from a decrease in the interest rates, he
will also suer a big loss if rates of interest increase. More precisely, let us
notice (for those readers who are statistically inclined) that the relation-
ship between dBaB and i,
dB
B
= D
m
di
is a linear one between the two random variables dBaB and di. The vari-
ance of (dBaB) is therefore,
VAR(dBaB) = D
2
m
VAR(di)
(9)
and therefore the standard deviation of dBaB is
o(dBaB) = D
m
o(di)
(10)
From (10), we can see immediately that the standard deviation of the
relative change in the bond's price is a linear function of the standard
deviation of the changes in interest rates, the coecient of proportionality
being none other than the modied duration.
4.6 Properties of duration
We will now review the three main properties of duration, namely the
relationship between duration and
.
coupon rate
.
yield to maturity
.
maturity
4.6.1 Relationship between duration and coupon rate
Suppose that the coupon rate of a bond increases, the yield and maturity
remaining constant. The eect of this change in coupon rate on duration
can be analyzed from two points of view. First, we can make use of the
physical interpretation of duration, which we mentioned earlier, as the
weighted average of the discounted cash ows. We can immediately see,
for instance, from gure 4.1, that doubling the coupons will double all the
rst n 1 cash ows, but will not double the last one (because it contains
not only the coupon, but also the par value, which remains unaected by
the change). Therefore, the distribution of the weights will be tilted to the
left, and the center of gravity will also move to the left; the duration of the
bond will decrease. This result can be veried by considering the expres-
sion of duration in closed form. We can now show that duration in such a
closed form can be expressed as:
D = 1
1
i

T(i caB
T
) (1 i)
(caB
T
)[(1 i)
T
1[ i
(11)
Indeed, it is often convenient to use a closed form for duration, as
opposed to an open form using the sign
, in the same way that closed

forms for the value of a bond are often more useful than open ones. To
obtain this closed form for duration, we will make use of the fact that it is
equal to
D =
1 i
B
dB
di
and that the bond's value is
B =
c
i
[1 (1 i)
T
[ B
T
(1 i)
T
Dividing both sides by B
T
, we get
B
B
T
=
caB
T
i
[1 (1 i)
T
[ (1 i)
T
which can be written as the following product:
B
B
T
=
1
i
(caB
T
)[1 (1 i)
T
[ i(1 i)
T
80 Chapter 4
Taking the logarithm, we have
log(BaB
T
) = log i log(caB
T
)[1 (1 i)
T
[ i(1 i)
T
The derivative with respect to i is

d log(BaB
T
)
di
=
d log B
di
=
1
B
dB
di
=
1
i

(caB
T
)T(1 i)
T1
(1 i)
T
i(1 i)
T1
(T)
(caB
T
)[1 (1 i)
T
[ i(1 i)
T
Duration D is
(1i)
B
dB
di
. Multiplying the expression just above by
(1 i), we nally get expression (11) after simplications.
We can check if the coupon rate caB
T
increases, whether there will be a
denite decrease in duration, because the numerator of the far-right frac-
tion in (11) will decrease, and the denominator will increase. (As a word
of caution, this result cannot be obtained as easily from the open-form
expression of duration (2), because whenever the coupon rate changes, the
value of the bond B changes too; further analysis is then in order. This is
precisely the reason why it is simpler to use the closed-form of duration
(11) as we have done.)
4.6.2 Relationship between duration and yield
Should the yield to maturity increase, we can surmise from the diagram
picturing the center of gravity of the bond's cash ows that the masses
will be relatively more alleviated on the right-hand side of the diagram
than on the left-hand side. This must be so, because the discount factor is
(1 i)
t
; changing i will modify relatively more the high-order terms
(those with high t's) than the low-order ones. Therefore the center of
gravity will move to the left and duration will be reduced. Of course, this
property needs to be conrmed analytically.
This time, the easiest way to proceed is to take the derivative of the
open-form of duration (2) with respect to i. We get a result that will be
most useful in our chapter on convexity. The result is
dD
di
= (1 i)
1
S (12)
where S is the dispersion, or variance of the payment times of the bond,
and as such is always positive. Hence
dD
di
is always negative, and duration
decreases with an increase in yield to maturity.
Let us now establish this result. From the expression of duration
D =
1
B
T
t=1
tc
t
(1 i)
t
we can write
dD
di
=
1
B
2
T
t=1
t
2
c
t
(1 i)
t1
B(i)
T
t=1
tc
t
(1 i)
t
B
/
(i)
_ _
Factoring (1 i)
1
and multiplying each term in the bracket by 1aB
2
, we
get
dD
di
= (1 i)
1
_
T
t=1
t
2
c
t
(1 i)
t
B(i)
(1 i)
B
/
(i)
B(i)
T
t=1
tc
t
(1 i)
t
B(i)
_
_
The second term in the bracket is simply D
2
. We may thus write
dD
di
= (1 i)
1
_
T
t=1
t
2
c
t
(1 i)
t
B(i)
D
2
_
_
We recognize in the bracket the variance, or dispersion, of the times of
payment t; indeed, it is equal to the weighted average of the squares of the
dierences between the times t and their average D. For simplicity, let us
denote
w
t
=
c
t
(1 i)
t
B(i)
so that
T
t=1
w = 1, and D =

T
t=1
tw
t
. The bracket then becomes equal to
T
t=1
t
2
w
t
D
2
=
T
t=1
t
2
w
t
2D
T
t=1
tw
t
D
2
T
t=1
w
t
=
T
t=1
w
t
(t
2
2tD D
2
) =
T
t=1
w
t
(t D)
2
82 Chapter 4
and this last term is none other than the variance of the times of payment,
which we will denote as S. Therefore equation (12) is true.
In the case of duration as well as in the case of convexity, the weights w
t
are just the shares of the bond's cash ows (in present value) in the bond's
value. Of course, the dispersion S is measured in years
2
and is always
positive. We have thus proved that dDadi is always negative.
4.6.3 Relationship between duration and maturity
Contrary to our intuition and to what some supercial analysis might lead
us to believe, there is not always a positive relationship between duration
and maturity, in the sense that an increase in maturity does not necessar-
ily entail an increase in duration. Let us rst state the results.
.
Result 1. For zero-coupon bonds, duration is always equal to maturity.
Therefore, duration increases with maturity.
.
Result 2. For all coupon-bearing bonds, duration tends toward a limit
given by 1
1
i
when maturity increases innitely. This limit is independent
of the coupon rate.
.
Result 3. For bonds with coupon rates equal and above yield to maturity
(equivalently, for bonds above par) an increase in maturity entails an in-
crease in duration toward the limit 1
1
i
.
.
Result 4. For bonds with coupon rates strictly positive and below yield
to maturity (equivalently, for bonds below par), an increase in maturity
has the following consequences: duration will rst increase; it will pass
through a maximum; and then it will decrease toward the limit 1
1
i
.
If result 1 is obvious, results 2, 3, and 4 are quite surprising, to say the
least. Let us tackle these in order.
In result 2, the duration of all coupon-bearing bonds tends toward a
common limit. This can be seen from the expression of duration in closed
form (11); indeed, the numerator in the large fraction on the right-hand
side of (11) is an ane function of T, and the denominator of the same
fraction is a positively-sloped exponential function of T. When T goes to
innity, the exponential will ultimately predominate over the ane, and
the fraction will go to zero, letting duration reach a limit equal to 1
1
i
.
(For example, if i = 10%, duration tends toward 1
1
0X1
= 11 years.) This
surprising result, together with its delightful simplicity, compels us to seek
an explanation. Why does duration tend toward 1
1
i
when the maturity
of the bond, whatever its coupon, goes to innity? We suggest the fol-
lowing interpretation, which permits us to get to this result without
recourse to any calculation.
We merely have to think about two conceptsduration and the rate of
interestand see how they are related. Duration can be thought of as the
average time you have to wait to get back your money from the bond
issuer. The rate of interest is the percent of your money you get back per
unit of time; hence the inverse of the rate of interest is the amount of time
you have to wait to recoup 100% of your money. Thus, you will have to
wait
1
i
years to recoup your money; however, since payment of interest is
not continuous but discontinuous (you have to wait one year before
receiving anything from your borrower), you will have to wait 1 year
1
i
years, which is exactly the limit of the duration when maturity extends to
innity. (The reader who is curious enough to evaluate a bond with a
continuously paid coupon will nd that its duration is
1
i
, just as foreseen.)
Results 3 and 4 reect the ambiguity of the eect of an increase of
maturity on duration. For large coupons (for caB
T
b i), duration
increases unambiguously, but for small coupons, duration will rst in-
0 10 20 30 40 50 60 70 80 90 100 110
0
50
40
30
20
10
D
= 11
Maturity (years)
D
u
r
a
t
i
o
n

(
y
e
a
r
s
)
0.3%
Zero-coupon
5%
15%
10%
7%
Figure 4.2
Duration of a bond as a function of its maturity for various coupon rates (i F10%).
84 Chapter 4
crease and then decrease. In order to understand the origin of this ambi-
guity, our analogy with the center of gravity will prove helpful once more.
Let us rst see why our intuition leads us in a false direction. If matu-
rity increases by one year, we are not always adding some weight to the
right end of our scale, as we might be tempted to think. What we are in
fact doing is increasing the right-end weight by adding at time T 1, in
present value, the par value B
T
plus an additional coupon; but we are
alleviating the same end since the present value of par B
T1
is smaller
1 T1

T T+1
1 T1

T
A
A
A'
1 T1 T
1 T1 T T1
Adding one additional cash ow, at T 1, has two eects on the right
end of the scale:
1. It adds weight at the end of the scale, rst by adding a new coupon at
T 1, and second by increasing the length of the lever corresponding to
the reimbursement of the principal.
2. It removes weight at the end of the scale by replacing A with a smaller
quantity, A
/
= Aa(1 i).
Therefore the total eect on the center of gravity (on duration) is ambig-
uous and requires more detailed analysis.
Figure 4.3
An intuitive explanation of the reason why the relationship between duration and maturity is
ambiguous.
than the present value of B
T
at time T (see gure 4.3). Therefore, we are
not sure what the nal outcome of this process will be. We can easily
surmise, however, that if the coupon is small relative to the rate of inter-
est, we will remove some weight from the right end of the scale, and
therefore the center of gravity will move to the left (the duration will
decrease)which is exactly what happens.
This is a good example of where further analysis is required. Strangely,
only in 1982 was the analysis of the relationship between duration and
maturity completely carried out (by G. Hawawini).6 The derivations
for results 3 and 4 are unfortunately rather tedious to obtain7 and are
not given here; instead we refer the interested reader to their original
source.
4.7 Duration of a bond portfolio
Let us now consider the duration of a portfolio of bonds. First, for sim-
plicity, restrict the number of bonds to two, whose prices are denoted
B
1
(i) and B
2
(i), respectively. Consider a portfolio of N
1
bonds ``1'' and
N
2
bonds ``2''.
We remember that the duration of a coupon-bearing bond B
i
was just
D =

T
t=1
tc
t
(1 i)
t
aT, and we showed it was equal to
1i
B
dB
di
. It is
only natural to treat a single coupon-bearing bond as a portfolio of
bonds; a portfolio P of coupon-bearing bonds is of course such that its
duration, denoted D
P
, will be equal to
1i
P
dP
di
. Let us therefore determine
this last expression.
The portfolio's value is
P = N
1
B
1
(i) N
2
B
2
(i) = P(i)
Taking the derivative of P(i) with respect to i, we get
dP(i)
di
= N
1
dB
1
(i)
di

N
2
dB
2
(i)
di
6G. Hawawini, `Òn the Mathematics of Macauley's Duration,'' in: G. Hawawini, ed., Bond
Duration and Immunization: Early Developments and Recent Contributions (New York and
London: Garland Publishing, Inc., 1982).
7To make matters worse, in the case of the curves representing duration of under-par bonds
as a function of maturity, the abscissa of their maximum cannot be tracked with an explicit
expression and therefore has to be given implicitly.
86 Chapter 4
Multiply each member of the above equation by (1 i)aP. Then mul-
tiply and divide the rst term of its right-hand side by B
1
(i) and the
second term by B
2
(i), respectively. We get
1 i
P(i)
dP(i)
di
=
N
1
B
1
(i)
P(i)

1 i
B
1
(i)
dB
1
(i)
di
_ _
N
2
B
2
(i)
P(i)

1 i
B
2
(i)
dB
2
(i)
di
_ _
The above expression may then be written as
D
P
=
N
1
B
1
(i)
P(i)
D
1

N
2
B
2
(i)
P(i)
D
2
Denoting N
1
B
1
(i)aP(i) IX
1
and N
2
B
2
(i)aP(i) IX
2
(with X
1
X
2
= 1),
D
P
reduces to
D
P
= X
1
D
1
X
2
D
2
= X
1
D
1
(1 X
1
)D
2
and therefore the duration of a portfolio is just the weighted average of
the bond's durations, the weights being the shares of each bond in the
portfolio's value.
The extension of this rule to an n-bond portfolio is immediate, and we
have
D
P
=
n
j=1
X
j
D
j
Y X
j
I
N
j
B
j
(i)
P(i)
and
n
J=1
X
j
= 1
An important caveat is in order here. From a practical point of view, it
is incorrect to add or average durations (or modied durations) of bonds
that do not carry the same yield to maturity. Furthermore, we have dealt
only with obligations that do not carry any signicant default likelihood
(which is denitely not the case with some corporate securities, even less
with some bonds issued in emerging markets, either by governments or
rms). In these latter cases, duration has little to do with any signicant
measure of the investor's risk.
4.8 Measuring the riskiness of foreign currencydenominated bonds
Before leaving this chapter on duration and its properties, let us consider
the risk of foreign bonds. As the reader may guess, their risk (apart from
any default from the issuer) stems from the following sources: volatility of
the foreign interest rate, duration, and foreign exchange risk. It will be
interesting to see how these three kinds of risk are related and to assess
their relative importance.
Let B represent the value of a foreign bond (for instance, a Swiss bond
held by an American owner). Let e be the exchange rate for the Swiss
franc (number of dollars per Swiss franc). If V is the value of the Swiss
bond expressed in dollars, we have
V = eB (13)
Taking logs on both sides, and then the dierential, we get
d ln V = d ln B d ln e (14)
which may be written also as
dV
V
=
dB
B

de
e
(15)
All three relative increases involved in (15) are random variables; in linear
approximation the relative increase of the Swiss bond, expressed in dollars,
is the sum of the relative increase of the Swiss bond, expressed in Swiss
francs, and the relative increase of the Swiss franc. Should the Swiss franc
rise, for instance, this will increase the performance of the Swiss bond
from the American point of view.
If i designates the Swiss rate of interest, we have, using the property of
duration we have just seen (equation (7)),
dB
B
=
1
B
dB
di
di =
D
1 i
di (16)
and therefore we include both the duration of the foreign bond and pos-
sible changes in the foreign rate of interest in equation (15), which
becomes
dV
V
=
D
1 i
di
de
e
(17)
Let us now take the variance of the relative change in value of the foreign
88 Chapter 4
bond; we get
VAR
dV
V
_ _
=
D
1 i
_ _
2
VAR(di) VAR
de
e
_ _
2
D
1 i
COV diY
de
e
_ _
(18)
The covariance between changes in interest rates and modications in
the exchange rates is usually very low, and the bulk of the variance of
changes in foreign bonds' value stems mainly from the variance of the
exchange rate. Empirical studies8 show the share of the exchange rate
variance is easily two-thirds of the total variance; for instance, for an
American investor, this share was 61% for U.K. bonds, 86% for Swiss
bonds, and 94% for Belgian bonds.
An important caveat is in order here. We have underlined the fact that
equation (15) gives the relative change in the Swiss bond's value from the
point of view of an American investor, in linear approximation. This is
a perfectly acceptable approximation if the changes both in the bond's
value and in the exchange rate are small (on the order of a few percent).
However, it is not a good approximation if those changes are large, as
they often are in emerging markets.
Many practitioners fall into the trap of using this formula for interna-
tional investments in stocks or bonds even when those rates of changes
are large (for instance, 20% or 40%); they simply add up the relative
change in the domestic value of the investment with the relative change
in the exchange rate. Even a highly respectable and otherwise excellent
periodical (whose name we will not divulge) made this mistake when, in
1998, it published tables on the declines of emerging East Asian markets
from the point of view of an international investor. This even led them to
draw a diagram in which they stated that the loss on the Indonesian stock
market had been 123%. This result of course does not make sense; even if
you invest in Martian stocks, you cannot lose more than your investment,
8See the study by Steven Dym, ``Measuring the Risk of Foreign Bonds,'' The Journal
of Portfolio Management (Winter 1991): 5661. For more recent data on volatility and cor-
relation of bond markets, see Bruno Solnik, Cyril Boucrelle, and Yann Le Fur, `Ìnterna-
tional Market Correlation and Volatility,'' Financial Analysts Journal (September/October
1996): 1734.
which is 100%. Likewise, if you bet your shirt in Indonesia and lose it,
nobody will ask you to throw in your tie as well.
Where does the error come from? We will now show how the true loss,
in exact value, should be calculated.
As before, let B denote the domestic value of the international invest-
ment (in that case, the investment in Indonesian rupees), and let e desig-
nate the exchange rate (number of dollars per Indonesian rupee). Let
V = Be designate the value of the investment in Indonesia expressed in
dollars. Let us determine the relative rate of change (which, of course, can
be negative) in the investment's value V. Suppose that B changes by hB,
and e moves by he. We then have
hV
V
=
(B hB)(e he) Be
Be
=
Be ehB Bhe hBhe Be
Be
so, nally,
hV
V
=
hB
B

he
e

hB
B

he
e
(19)
So the rate of change in the international investment is exactly equal to
the sum of the relative change in the domestic investment plus the relative
change in the exchange (hBaB heae), to which you have to add the
product of the relative change of the domestic investment and the relative
change in the exchange rate, (hBaB) (heae).
An example will illustrate this. Let us come back to the case we just
mentioned. Suppose that the loss on the Jakarta stock market was 60% in
a given period and that the rupee lost 60% of its value against the dollar
in the same period. The rate of change in the investment's value from
the point of view of an American investor is of course not minus 120%.
It is
hV
V
= 60% (60%) (60%)(60%)
= 120% 36% = 84%
90 Chapter 4
This result is not only exact but makes good sense, as can easily be seen
in the following illustration. Suppose your little sister has just baked a
delicious chocolate cake in the shape of a rectangle. In the short time she
has her back turned, you rst slice o 60% of the length of the cake,
swallow it with delight, and cannot resist stealing another 60% of its
width. After your rst mischief, 40% of the cake is left; your second
prank has taken away 60% 40% = 24% of what remained, so that only
40% 24% = 16% is left. (Equivalently, you could say that 40% of
40%, that is, 16%, is left.) Your sister has lost indeed 84% of her cake.
You can, however, try to comfort her by explaining that, although you
were mischievousfor which you deeply apologizeyou did not take
away 120% of the cake, as many people would unjustly accuse you of
doing.
The good news is that the true value is easy to calculate; the bad news
is that the expected value, variance, and probability distribution of an
expression like (19) are very dicult to analyze. Hence the popularity of
the linear approximation we mentioned, which is valid, however, only for
small changes in asset values and exchange rates.
Key concepts
.
Duration is the weighted average of the times to payment of all the cash
ows generated by the bond; the weights are the cash ows in present
value.
.
Duration can be viewed as the center of gravity for all the bond's dis-
counted cash ows, if these are set on a horizontal time axis.
.
An important property of duration is the bond's sensitivity to interest
rate risk; more fundamentally, duration is at the heart of the immuniza-
tion process (a method of protecting an investor against uctuations in
the rate of interest).
.
Duration decreases with coupon rate and interest rate; it always in-
creases with maturity for above-par bonds (i.e., when the coupon rate is
above the interest rate). For below-par bonds, except zero-coupon bonds,
duration rst increases with maturity and then decreases.
.
In all cases except for zero-coupons, duration tends toward a limit when
maturity increases to innity.
Basic formulas
.
Duration = D =

n
t=1
tc
t
(1 i)
t
aB = 1
1
i

T(icaB
T
)(1i)
(caB
T
)[(1i)
T
1[i
.
Duration as a measure of the sensitivity of the bond's price to interest
rate (or yield to maturity):
D =
1i
B
dB
di
.
Modied duration ID
m
=
D
1i
=
1
B
dB
di
.
Relation between duration and yield:
dD
di
= (1 i)S
where S is the variance (the dispersion) of the bond's times of payment
.
Limit of duration when T yX
1
i
1.
.
Duration of bond portfolio (valid only for bonds of same yield to ma-
turity);
D
p
=

j=1
X
j
D
j
.
Relative rate of increase in value of a foreign-denominated bond:
hV
V
=
hB
B

he
e

hB
B

he
e
where e is the exchange rate.
Questions
4.1. Consider equation (6) of section 4.4. In which units are both sides of
the equation measured?
4.2. What are the measurement units of modied duration?
4.3. Conrm that equation (11) in section 4.6.1 is exact.
4.4. In each of the following cases, what is the duration of a bond with
par value: $1000; coupon rate: 9%; maturity: 10 years?
a. i = 9%
b. i = 8%
Comment on your results.
92 Chapter 4
4.5. Two bonds, A and B, have the following characteristics:
Bond A Bond B
Maturity 15 years 7 years
Coupon rate 10% 10%
Par value $1000 $1000
Suppose that the rate of interest increases from 12% to 14%. Determine
the relative change in the bond's values, and comment on your results.
The relative change in the bond's value should be determined in exact
value (and denoted hBaB) and linear approximation (denoted dBaB).
Will these relative changes be more pronounced in exact value or in linear
approximation?
4.6. Consider question 4.5 again, this time using the following data:
Bond A Bond B
Coupon rate 12% 8%
Par value $1000 $1000
Comment on your results, which you can indicate in exact value or in
linear approximation.
4.7. Consider question 4.5 again, this time using the following data:
Bond A Bond B
Coupon rate 10% 14%
Par value $1000 $1000
Comment on your results. How can you compare these results to those
you obtained in the original question 4.5?
4.8. Repeat question 4.5, using the following data:
Bond A Bond B
Coupon rate 10% 5%
Par value $1000 $1000
4.9. Finally, consider question 4.5 with the following data:
Bond A Bond B
Coupon rate 2% 2%
Par value $1000 $1000
What strikes you in your results?
4.10. Verify that equation (18) follows from equation (17). (See section
4.8.)
94 Chapter 4
5
DURATION AT WORK: THE RELATIVE
BIAS IN THE T-BOND FUTURES
CONVERSION FACTOR
Dion. For present comfort, and for future good . . .
The Winter's Tale
King Lear. . . . that future strife
May be prevented now.
King Lear
Chapter outline
5.1 Forward and futures markets: a brief survey 96
5.1.1 The forward contract and its deciencies 96
5.1.2 Introducing futures markets 99
5.1.3 Futures trading 100
5.1.4 The actors in futures markets 100
5.1.4.1 Hedgers 100
5.1.4.2 Speculators 101
5.1.4.3 Arbitrageurs 101
5.1.5 The organization of safeguards: the role of the clearinghouse
and the margin requirements 102
5.1.6 The system of daily settlements and margins 102
5.1.7 The riskiness of a ``naked'' futures position 106
5.1.8 Closing out a futures position 107
5.1.8.1 Delivery 107
5.1.8.2 Cash settlement 109
5.1.8.3 Osetting positions or reversing trades 110
5.1.8.4 Exchange of futures for physicals (EFP
agreement) 110
5.1.9 Consequences of failure to comply 111
5.2 Pricing futures through arbitrage 111
5.2.1 Convergence of the basis to zero at maturity 112
5.2.2 Properties of the basis before maturity, F(tY T) S(t):
introduction 112
5.2.3 A simple example of arbitrage when the asset has no carry
cost other than the interest 116
5.2.4 Arbitrage with carry costs 117
5.2.4.1 The case of an overvalued nancial futures
contract 121
5.2.4.2 The case of an undervalued nancial futures
contract 122
5.2.5 Possible errors in reasoning on futures and spot prices 123
5.3 The relative bias in the T-bond futures contract 125
5.3.1 The various types of options imbedded in the T-bond
futures contract 125
5.3.1.1 The wild card (time to deliver) option 125
5.3.1.2 The quality option 126
5.3.2 The conversion factor as a price 127
5.3.3 The ``true'' or ``theoretical'' conversion factor 128
5.3.4 The relative bias in the conversion factor 129
5.3.5 The prot of delivering and its biases 134
Appendix 5.A.1 The calculation of the conversion factor in practice 139
Overview
As we will show in this book, duration has many practical applications.
Perhaps one of the most surprising is its application as a measure of the
relative bias that is introduced in one of the most important nancial
contracts on organized markets, the 30-year Treasury bond futures con-
tract, set up by the Chicago Board of Trade (CBOT). Before showing
this, we will review briey what forward and futures markets are about
and then describe the various futures contracts on bonds. This is a long
chapter, and the rst section can be skipped by the reader who is already
familiar with forward and futures markets.
5.1 Forward and futures markets: a brief survey
Before examining the characteristics of a futures contract, it is important
to review those of the simpler forward contract. We will thus see how the
necessity of setting up a system of futures contracts stems in a natural way
from some drawbacks of the forward contract.
5.1.1 The forward contract and its deciencies
Uncertainty has important consequences for all agents who have to make
transactions in the future. Consider, for instance, a producer of wheat (a
farmer) and a user of wheat (a mill). Each of those agents incurs a risk
with regard to the price at which he will exchange wheat in the future.
Suppose each agent is willing to secure a xed price, set today, for the
exchange of wheat that will take place in, say, six months. That price can
be denoted p(tY T), the forward price set today (at time t) for a transac-
tion that will occur at time T. In such a case we say that the farmer and
96 Chapter 5
the mill enter into a forward contract to exchange a commodity of certain
specications, in a given amount, for a price p(tY T). Without any hedg-
ing, the farmer runs the risk of having to sell at a price below p(tY T); on
the other hand, he might make an unexpected prot if the spot price at
time T, denoted S(T), is above p(tY T). The situation is symmetrical for
the mill: it runs the risk of having to pay more than p(tY T), and it may
also cash in on an unexpected prot if the price happens to be lower than
p(tY T).
Suppose that each of those agents is willing to forego the possible ad-
vantage that uncertainty may bring about by securing a transaction at
p(tY T).
Consider rst the farmer's position. He will be holding the wheat and
is willing to sell it; he is said to have a short position (and he is called
the short). Figure 5.1 represents the farmer's possible payos at time T,
according to every possible future spot price S
T
. His proceeds will be, in
terms of price, the 45
line (equal to S(T)). He incurs considerable risk

if the price is very low. Suppose now that at time t he enters a forward
contract (he takes a so-called short position on each unit of wheat) at
p(tY T). He therefore has the obligation of selling one unit of wheat at
Payoff
p(t,T)
p(t,T) S(T)
(a) Delivering crop
at price S(T)
0
(c) Entering a short forward
contract and delivering crop:
p(t,T) S(T) + S(T) = p(t,T)
(b) Entering a short forward
contract: p(t,T) S(T)
Figure 5.1
Possible monetary outcomes for the farmer (the short).
97 Duration at Work
time T at a price p(tY T) established at time t. What will his payo be if he
does not own the wheat? Suppose for the time being that anyone entering
such a contract has to deliver one unit of a certain quality of wheat at
time T. His payo will be p(tY T) minus the cost of acquiring this unit,
that is, the spot price S(T); it will thus be p(tY T) S(T). The payo for
this contract will be the negatively sloped line going through p(tY T) on
the S
T
axis. Indeed, if at time T the spot price S(T) is lower than p(tY T),
he receives the positive amount p(tY T) S(T); if, on the contrary, the
spot price is higher than p(tY T), he loses S(T) p(tY T) (or receives
[S(T) p(tY T)[ = p(tY T) S(T), a negative amount). In any case, he
always receives S(T) p(tY T), an amount that may be positive or
negative. Now his total position, if at time T he owns the commodity and
at the same time receives the proceeds of his futures position, is simply
S(T) S(T) p(tY T) = p(tY T), the horizontal line at height p(tY T).
Indeed, he has now secured a comfortable position; no matter what the
future spot price may be, he will cash in p(tY T). He has bought this safety
by foregoing any possible benet he might have incurred if the future spot
case S(T) had been above p(tY T).
Let us now turn to the mill. If it does not enter a futures contract, it will
have to buy one unit at price S(T); its cash outow is represented by the
negatively sloped line S(T) going through the origin (see gure 5.2).
This entails incurring the risk of having to buy wheat at a very high price,
thus disbursing a high amount S
T
. On the other hand, suppose the mill
takes a long position at p(tY T); the payo from this futures contract is the
positively sloped, 45
line going through p(tY T), equal to S(T) p(tY T).

The total payo, corresponding to buying the wheat and buying the
futures contract, is simply S(T) S(T) p(tY T) = p(tY T). So the
mill is going to incur a cash outow of p(tY T), no matter what happens
on the wheat spot market at time t. The mill pays this security by fore-
going the possibility of buying wheat at lower prices than p(tY T) if the
market spot price drops below p(tY T) at time T. Notice also that, both
agents being considered together, the disbursements of one agent (the
mill) will be equal to the receipts of the other (the farmer). This is basi-
cally the mechanism of a forward contract.
However, many problems would arise with such a contract, the most
important one being the possibility of default by one of the parties. A
second problem is that neither party would be sure of obtaining the best
possible deal by entering into this agreement; the seller might well think
98 Chapter 5
he should have looked around among other potential buyers to see if he
could have received a better price, while the buyer could also regret not
having shopped around more.
A third drawback of a forward agreement is that no party can change
his mind during the period of the contract. For instance, should the short
party (the party with the obligation to sell) change his mind, the only
possibility would be for him to try to sell his contract to a third party, and
this could be very costly.
The reader should not infer from the above that forwards can't trade.
On the contrary, they do trade in large volumes. This comes from the fact
that standardized contracts (as futures contracts are, as we will see) may
be not exible enough for the participants' wishes. Also, liquidity may be
higher in some forward deals. For instance, there is a huge volume of
bank-to-bank over-the-counter deals, where liquidity can be higher than
in exchanged traded contracts like futures.
5.1.2 Introducing futures markets
As the reader will notice, each of the three major drawbacks of a forward
contract carries a given type of risk: the risk of default, the risk of making
the transaction at an over- (or under-) stated price, and the risk of a party
Payoff
p(t,T)
p(t,T)
S(T)
(b) Entering a long forward
contract: S(T) p(t,T)
0
(c) Entering a long
forward contract and
buying the crop: p(t,T)
(a) Buying the
crop: S(T)
Figure 5.2
Possible monetary outcomes for the mill (the long).
99 Duration at Work
bearing a large loss if that party wants to change its mind. And such a
contract was originally intended to remove risk from the transaction.
For these reasons, and others that we will briey review, forward mar-
kets have been progressively developed into organized ``futures'' markets,
the main features of which are the following: (a) futures contracts are
traded on organized exchanges (a clearinghouse makes certain that each
party's obligation is fullled); (b) each contract is standardized; (c)
through a system of margin payments and daily settlement, the position is
cleared every day, as we will explain.
5.1.3 Futures trading
In the United States, the rst organized futures exchanges were trading
grains in the nineteenth century; other categories of commodities came to
be traded on such exchanges. Today, soybeans, livestock, energy products
(from heating oil to propane), metals, textiles, forest products, and food-
stu futures are traded on such exchanges as the Chicago Board of Trade
or the New York Mercantile Exchange. Financial instruments were rst
traded on futures markets in the 1970s, and they have come to play a
crucial role. It has become customary to distinguish three kinds of nan-
cial futures: interest-earning assets, foreign currencies, and indices. In this
text we will be interested primarily in interest-earning assets. Note that
the common wording for futures on those assets is interest rate futures,
although the traded object is of course the underlying nancial asset (for
example, a short-term (one year) Treasury bill or a long-term (20 year)
T-bond).
5.1.4 The actors in futures markets
It has become customary to classify the operators in futures markets into
three groups: hedgers, speculators, and arbitrageurs.
5.1.4.1 Hedgers
In our introduction, we dealt at length with the hedging motive some
agents may have. Some have good reason to hedge against a low future
price (those who own the commodity or who will own itperhaps be-
cause they will be producing it). Others hedge against a high future price
(those who will have to buy the commodity, presumably because it enters
their own chain of production).
100 Chapter 5
Also, as we will see later, it is quite possible to hedge against mishaps to
a given commodity while entering a futures market of a slightly dierent
commodity, if the two commodities happen to be suciently related that
their prices are highly correlated. One refers to such transactions on the
futures market as cross-hedging.
5.1.4.2 Speculators
``To speculate'' comes from the Latin speculare, ``to forecast.'' The Latin
word also conveys an overtone of divination, especially in the noun
``speculatio.'' Many agents try to forecast what the future spot price of
any commodity will be and will compare their forecast to the futures price
being established on the futures market. Those agents, who do not hold
the commodity and do not intend to hold it, take considerable risk. They
can very quickly lose their money if shorts see the futures price climb or if
longs witness a plunge. Of course, their risk is rewarded by substantial
prots if, on the contrary, things turn out their way. In order to make
forecasts, some will rely on fundamental analysis; they will try to forecast
future general conditions for the supply and demand of the commodity
and infer an expected spot price. If the dierence between that estimate
and the futures price is large enough, they will take a position on the
futures market. In any case, of course, spot and futures prices are funda-
mentally driven by the same forces, and there is a denite, strong rela-
tionship between the two prices, which we will examine very soon.
Positions held by speculators can be held for variable periods: from a
few months for long-term speculators to a few seconds for the so-called
``scalpers'' (individuals who take bets on the fact that sell orders, for in-
stance, may well be soon followed by buy orders). Scalpers prot from a
small dierence between the price at which they sell the buy orders and
the price at which they bought the sell orders.
5.1.4.3 Arbitrageurs
Arbitrageurs are traders who prot from mismatches between spot and
futures prices and between futures prices with dierent maturities. In
order to give some examples of such arbitrages, in the next section we
discuss the all-important question of the determination of futures prices.
We should stress now that we will show what the links are between spot
and futures prices today, at a given point in time; of course, we are not
able to know what the spot or the futures prices will be at any distant
point in time.
101 Duration at Work
5.1.5 The organization of safeguards: the role of the clearinghouse
and the margin requirements
Contrary to a forward contract, where each party has to organize itself
against possibilities of default from its counterpart, exchanges set them-
selves between the parties; each party deals separately with the exchange.
Also, in contrast to a forward agreement, the specications of a futures
contract are always standardized: the unit amount of the contract, the
exact quality of the product to be traded, and the time and location of
delivery are all set in a unique and precise way. In addition, the exchange
will not generally permit daily uctuations of the futures price that would
exceed certain limits.
Each futures market has a clearinghouse that is closely associated with
it, either a part of the exchange or a separate corporation. In either case,
the clearinghouse buys the futures contract from the short party and sells
it to the long. Thus the short and the long are obligated only to the
clearinghouse (through their respective brokers). Also, the clearinghouse
runs practically no risk for the following two reasons. First, for each trade
it makes with any given party, it makes the opposite trade with another
party, and thus holds no net (positive or negative) position; second, it
requires from each trading party security deposits (or margins) to ensure
that if ever one party defaulted its obligations to the clearinghouse, it
would be covered against that kind of default. We will now explain what
kind of margins are required both from the short and the long.
5.1.6 The system of daily settlements and margins
Anyone entering a futures agreement commits himself at time t to buy or
deliver a given commodity or a nancial product at some point of time T
in the future, for a given futures price F(tY T).1 The system of daily set-
tlements and margins, which we will now describe, is a very ecient way
to make sure each participant in the market will indeed honor its obliga-
tions, on the one hand, and on the other enable each participant to end its
commitment at any point in time.
For clarity, let us go back to the case of a forward contract. Suppose
you set up a system of guarantee deposits by which you make sure that
any participant in a forward market will fulll its obligations. How much
1 Throughout this book, we denote a futures price as F(tY T) and a forward price as p(tY T).
102 Chapter 5
would you require as a guarantee from an individual A who, for instance,
commits himself at time t to sell one unit of a given commodity at a cer-
tain time T, at price p(tY T)? At time T, the spot price of the commodity
S(T) may very well be under p(tY T) or above it. In the rst case, A earns
the dierence between the two. If A is a speculator, he can buy the com-
modity at T, pay S(T), sell it at the higher price p(tY T), and pocket the
dierence. If he owns the commodity, he also receives the dierence
p(tY T) S(T). The risk of default arises if S(T) is above the forward
price p(tY T), because the speculator or the owner of the commodity will
incur a loss on that contract equal to S
T
p(tY T). Indeed, the speculator
may simply not buy the commodity at price S(T) and sell it at p(tY T),
and instead disappear. The opposite risk appears in the case of party B,
who has committed himself to buying the commodity at price p(tY T);
should S(T) be lower than the agreed on price p(tY T), B will certainly not
be pleased at the prospect of buying a commodity at a price higher than
that of the spot market, since he will incur a loss p(tY T) S(T) when he
sells it. Also, if B is a user of the product, he may very well want to buy it
on the spot market instead of fullling his commitment to buy it at the
higher price p(tY T).
You observe that the maximum loss that any of the two parties A and
B can incur is not the whole forward price p(tY T), but only the dierence
between p(tY T) and S(T) in absolute value. Therefore you will require
from both parties a guarantee that will be in the order of magnitude of
this possible dierence, which you do not know at time t and which you
will have to forecast for date T. Furthermore, you will want to resettle the
contracts at regular intervals between t and T, perhaps every day, so that
each party has an easy way to exit the market. Finally, you observe that
the required guarantee will certainly diminish if it has to cover default risk
over a much shorter period (one day) than T t (which may be nine
months). The market will become more liquid, and fewer funds will be
tied up in the guarantee process.
Thus you will establish futures markets, such that every participant has
to permanently meet a minimum safeguard called a maintenance margin.
In no circumstances can the deposit fall lower than that margin. (Note
that each exchange establishes a minimum maintenance margin, but
stockbrokers can set higher minimal values for some clients whom they
suspect to be more likely to default their obligations than others.) Above
this, clearinghousesor the stockbrokerswill require from their clients
an amount that will permit them to settle immediately any daily loss. This
is why, when a futures contract is initiated, each party will be required to
make an initial deposit, called the initial margin, which will be above the
maintenance margin by an amount of the order of about one third. (The
maintenance margin will therefore be about three fourths of the initial
margin.) The funds deposited may by posted in cash, in U.S. Treasury
Bills, or in letters of credit; of course, some of this collateral may earn
interest, which remains in the owners' hands. Should the trader's account
exceed the initial margin because daily settlements have been favorable to
the trader, he has the right to withdraw at any time the amount exceeding
the initial margin. On the other hand, should the account fall to the level
of the maintenance margin (or below), the trader will immediately receive
a call to replenish his account up to the level of the initial margin. Should
he not comply with this demand, he will have to face dire consequences,
which we will detail (in section 5.1.9) after we have explained how, nor-
mally, a trader can exit the futures market.
We now present an example of such a system of daily settlements. We
consider the case of a trader who, early on Tuesday, July 1, buys a Sep-
tember gold futures at price F(tY T) = $400 an ounce, and we trace the
evolution of his long position through the next nine trading days, ending
on July 11, according to the daily evolution of the price of the September
futures contract for gold. We suppose that one futures contract on gold
entails the obligation to buy (or sell) 100 ounces of gold at futures price
expressed in dollars per ounce of gold. The initial margin is set at $2000
and the maintenance margin at $1500. This implies that in all likelihood,
the daily variation is not expected to be above $500 per contract, or $5
per ounce.
On the rst day, when our buyer and our seller enter their contracts
with the exchange, they are asked to deposit $2000 each as initial margin
requirement. At the end of the rst day, suppose that the settlement price
turns out to be $405. (In fact, prices are quoted to the cent in this con-
tract; for simplicity of exposition, we limit ourselves to dollars.) The
September futures price is now $405. Our long buyer benets from this
increase in the futures price; now operators on the market are willing to
buy gold at $405, while he is the owner of a contract specifying a price of
$400. His contract is thus worth $5 100 = $500. In a way, if the con-
tract were of the ``forward'' type, he could sell it for $500; indeed, any-
body wanting to buy gold at the futures maturity date for $405 would be
104 Chapter 5
indierent in entering the market with a $405 commitment, or paying $5
for a $400 commitment, which is what it would amount to if he bought
for $5 the initial contract set at $400.
The clearinghouse then credits the long with $500, taken from the
account of the unfortunate short, who had contracted to sell 100 ounces
of gold (one contract) at the futures price of $400, which is now valued
at $405. If his contract were a forward contract, he would have to pay
someone $500 to buy his contract. If the short's position is ``marked to the
market,'' he is considered as having lost $5 per ounce, and he suers a
cash outow of $5 100 = $500. Having been so adjusted, both the short
and the long positions are considered cleared, and both traders could de-
part the futures market. If, for instance, the short stays in it, he will now
be considered to have a short position on that September contract, whose
maturity is now one day closer, with a price F(t 1Y T) = $405. Our
subsequent example refers to the cash ows of the long trader; for the
short trader, all those cash ows correspond to cash ows of identical
magnitude but opposite sign.
At the end of the rst day, the long has a cumulative gain of $500 and a
nal cash balance of $2500 (column 5 of table 5.1), which is the next day's
(t 1) initial balance (column 2). The next day, things go the long's way
again: the futures price rises to $407; he is credited with a cash inow of
$200; his cumulative gain is $700; and his nal cash balancethe initial
cash balance for t 2is now $2700. Unfortunately for our long, the
Table 5.1
The futures buyer's position; initial futures price: F(t, T)F$400//ounce on July 1
(1) (2) (3) (4) (5) (6)
Date
Settle-
ment
price
Initial
cash
balance
Mark to
market
cash ow
Cumu-
lative
gain
Final
cash
balance*
Mainte-
nance
margin call
7/1 405 2000 500 500 2500
7/2 407 2500 200 700 2700
7/3 401 2700 600 100 2100
7/4 394 2100 700 600 2000 600
7/7 392 2000 200 800 1800
7/8 391 1800 100 900 1700
7/9 387 1700 400 1300 2000 700
7/10 390 2000 300 1000 2300
7/11 391 2300 100 900 2400
* This nal cash balance takes into consideration margin calls, if any.
futures price declines on day t 3 to $401, entailing a cash ow of
$600, therefore bringing his cumulative gain to only $100. Thus the
nal cash balance is $2100. On the fourth day, the settlement price falls to
$394. This means a loss of $700 for our long. If that sum were taken from
his cash account, its balance would fall to $2100 $700 = $1400. This
amount is below the maintenance margin (which was set at $1500), and
such a situation triggers a maintenance margin call whose purpose is to
bring back the cash balance to $2000; therefore, the amount of the main-
tenance margin call is $600. Over the next three days the long's situation
deteriorates, because on July 9 the price takes a dive to $387; the cash
balance at the beginning of that day was $1700; it is depleted now by a
loss of $400 and would thus fall to $1300, below the $1500 maintenance
margin. This triggers a second maintenance margin call in the amount of
$700.
In the last two trading days the long makes up a little for his losses,
because the futures price has risen somewhat.
At the end of trading on July 11, the long's situation is as follows. The
futures price is at $391. His loss can be determined in any of the following
four ways:
1. (change in futures price) 100 = [F(t 8Y T) p(tY T)[ 100 = (391
400)100 = 900
2. sum of mark to market cash ows = 900 [sum of column (3)]
3. cumulative net gain = 900 [last entry of column (4)]
4. nal cash balance [last entry of column (5)] minus initial cash balance
[rst entry of column (2)] minus all maintenance margin calls = 2400
2000 600 700 = 900.
5.1.7 The riskiness of a ``naked'' futures position
Notice how extremely risky any such futures position is,2 be it a long's
position or a short's. Our long has invested `ònly'' the $2000 initial
margin; he has lost $900 after nine days of trading. Had the futures price
been $360, corresponding to a 10% decline in the futures price of gold, he
would have lost $4000, 200% of his investment! If you buy a commodity
2 Such a position, not being covered by the present or future holding of the underlying
commodity (gold in this case), is said to be a ``naked'' position. Here we suppose that neither
the short holds the commodity, nor does the long intend to buy the commodity at maturity
of the contract; thus, both are supposed to be speculators.
106 Chapter 5
or an asset on the spot market, the maximum you can lose is 100% of
your investment, whereas on a derivatives market such as the futures
market, you can lose many times your investment. In the example of the
gold futures market, let r be the (possible) rate of relative decrease of the
futures price between the time you bought the contract (t) and time t d.
We have: r = [F(tY T) F(t dY T)[aF(tY T); the new futures price is
F(t dY T), and your loss is [F(tY T) F(t dY T)[ 100 = rF(tY T)100.
Suppose the relative decrease of the futures price is 20% and F(tY T) is
$400. The long loses $8000; as a percentage of his initial margin, he
loses
rF(tY T)100
2000
100 = 5rF(tY T)%. If the futures price declines by 50% and
F(tY T) is 400, he loses 1000% of his initial investment! In case of an in-
crease in the futures price, the same is true for the short, for whom r now
represents the rate of increase of the futures price, and who risks losing
5rF(tY T)% of his investment in only a few days of futures trading.
5.1.8 Closing out a futures position
It is now time to turn toward the all-important question of the means by
which a trader can voluntarily close out his position. As we will see, there
are basically four ways to accomplish this.
The vast majority of futures contracts do not result in actual delivery of
the underlying commodity or nancial product. Why is this so? For the
good reason that the majority of traders nd it much more convenient to
fulll their contracts through reversing trade, which simply osets their
positions. We will review all possibilities in turn, starting with the delivery
process. Indeed, it turns out that delivery may still be quite a rewarding
process, as is the case in the most important futures contract existing at
the time of this writing, the T-bond futures contract.
A trader can terminate a futures contract in four ways: (a) delivery, (b)
cash settlement, (c) reversing trade, and (d) exchange for physicals. We
will examine each in turn.
5.1.8.1 Delivery
Many futures contracts allow the short (the seller) to deliver the com-
modity in various grades and at various times within a time span (say, one
month). Why is this so? The futures exchanges have established such
provisions in order to facilitate the task of the short, by making the com-
modity's market more liquid. Indeed, if the contract was established so
that only a very short period (one day, for instance) was allowed for de-
livery and furthermore only a very specic kind or grade of the com-
modity could be delivered, such stringent conditions would have two
possible consequences. First, they could kill the market altogether, since
no one would be willing to commit himself to taking such a specic
commitment; second, those conditions could entice some to try to ``cor-
ner'' the market, by which we mean that an individual or a group of
individuals could engage in the following operation: they would buy on
the spot market an extraordinarily high proportion of the said grade of
the commodity, thus forcing the shorts to buy it back at a very high price
when obligated by their contract to make delivery.
Futures markets are regulated precisely to avoid such ``corners''; quite
a number of grades are deliverable, at dierent locations, over a rather
long time span. Also, any attempt to corner a market is a felony under the
Trading Act of 1921; civil and criminal suits can be launched against
traders who try to engage in such manipulations. Finally, the exchange
could force manipulators to settle at a xed, predetermined price. This is
what happened in the famous Hunt manipulation of the silver market at
the beginning of the 1980s.
More recently (in 1989), this also happened in a famous squeeze involv-
ing an important soybean processor. A long trader on the futures market,
the processor had also accumulated a large part of the deliverable soy-
beans. The squeeze started to drive prices up steeply, both on the cash and
the futures markets. Of course, the company's line of defense was to argue
that it was simply hedging its operations.3
Since only one futures price is quoted, while a set of deliverable grades
of the commodity are deliverable, exchanges have set up a system of
conversion factors whose purpose it is to equate, at least theoretically,
3 On this, see for instance: S. Blank, C. Carter, and B. Schmiesing, Futures and Options
Markets (Englewood Clis, N.J.: Prentice Hall, 1991), p. 384; and the articles they cite:
Kilman, S. and S. McMurray, ``CBOT damaged by last week's intervention,'' The Wall
Street Journal, July 1987, Sec. C, pp. 1 and 12; Kilman, S. and S. Shellenberger, ``Soybeans
fall as CBOT order closes positions,'' The Wall Street Journal, July 13, 1989, Sec. C, pp. 1
and 13. See also: Chicago Board of Trade, Emergency Action, 1989 Soybeans, Chicago,
Board of Trade of the City of Chicago, 1990.
We should add that, unfortunately, xed-income markets have not been exempt from
cornering attempts. For instance, a few bond dealers were accused in 1991 of cornering an
auction of two-year Treasury notes. On this, see S. M. Sunderasan, Fixed Income Markets
and Their Derivatives (Cincinnati: South-Western, 1997), pp. 67 and 69, as well as the refer-
ences contained therein, especially N. Jegadeesh, ``Treasury Auction Bids and the Salomon
Squeeze,'' Journal of Finance 48 (1993): 14031419.
108 Chapter 5
the attractiveness of delivering a particular grade. Consider rst, as an
example, the gold futures contract. Gold is available in various nesse
grades, and the conversion factor is simply the nesse percentage of each
grade. Will this distort the delivery process? If grade prices are exactly
proportional to nesseif the market just prices the gold content of any
gradethere will be no distortion, because the possible prot or loss of
any party remains as it should, proportional to the party's position.
However, this is not what happens in the gold market: prices are not
exactly proportional to gold content, and the short party, when making
delivery, will always have to account for the conversion factor and the
cost of acquiring gold to maximize his prot. Per unit of gold, this
prot will be equal to the quoted price times the conversion factor of the
delivered grade minus the cost of the delivered grade.
Before leaving this topic, we should underline that the futures contract
on Treasury bonds allows the delivery of quite a number of bonds and
that the conversion factors used by the Chicago Board of Trade (CBOT)
entail substantive dierences in prot for the short party. This issue is so
important that we will devote the third section of this chapter to it; there
we will derive a number of surprising results.
5.1.8.2 Cash settlement
The second way to close up a futures contract is through cash settlement.
This may occur if the contract is written on a nancial index or if a com-
modity is priced on an index. For instance, this is the case with the feeder
cattle contract of the Chicago Mercantile Stock Exchange, whereby a
cash settlement price is established as the average of prices of a number of
auctions held throughout the United States over the seven days preceding
settlement day. As another example, let us consider a long contract on
the most important U.S. stock index future: the Standard & Poor's 500
futures contract. The S&P Index is a value-weighted index of 500 major
stocks; as such it acts as the price of a basket of securities. One can set up
a contract on that basket, requiring from parties a future exchange of the
basket (the content of the index) for $450; of course, this contract would
be undeliverable because one could not trade fractions of stocks. It
would also be untradable because no one would ever sustain the costs of
acquiring so many stocks. Thus, it is quite natural that such a contract on
an index will be closed not through delivery but through settlement. In
fact, the minimum position on such a contract is the value of the index
times $500. The cash settlement the long would pay is the dierence
between the futures price he has agreed to pay and the spot value of the
index at maturity times $500. Suppose, for instance, that a party enters a
long futures contract at 400 and that the index falls to 350; he pays
400 $500 350 $500 = $25Y000. Had the index risen to 420, he
would have received 20 $500 = $10Y000.
5.1.8.3 Osetting positions or reversing trades
One of the reasons for organizing futures markets with contracts marked
to the market was to enable participants to enter and exit forward agree-
ments easily. Should any party want to exit a forward position early, he
would have to nd another party willing to trade the specic commodity
or nancial product for the initial settlement date, which might be cum-
bersome if not impossible. Also, the party would have to carry two for-
ward contracts (the long and the short) in his books until maturity.
Futures markets remedy both of these drawbacks at once by permitting
the long party wanting to get out of his futures contract to sell his long
position to another party. As a result, on the day the long party (A) wants
to get rid of his futures contract (which had been matched by a short, (B)),
he simply enters a short contract with the same specications as the initial
one. His new (short) position will now be matched with the long position
of a third party, called C. Party A has met all his obligations toward the
clearing house; he has received his gains or sustained his losses. Now C
has replaced A in its obligations toward B. By far the majority of futures
contracts are settled in this way because it is much less expensive and
often more convenient for all parties to operate in this way.
5.1.8.4 Exchange of futures for physicals (EFP agreement)
In the case where transportation of commodities is very costly (for
example, in the case of energy products like gasoline or heating oil), the
futures trading parties may try to negotiate delivery arrangements be-
tween themselves. Not only the location of delivery, but also its date and
the exact grade of the commodity, as well as its price, can be negotiated
between two parties. The parties may know each other, or the trade can
be facilitated by the exchange itself. The procedure is then as follows:
once all provisions of the agreement (including price) have been agreed
upon, A (the long, for instance) will sell his long contract to B (the short)
in exchange for the cash commodity. B does the same; through the ex-
110 Chapter 5
change, he will buy A's long contract in exchange for delivery of the cash
commodity. Both parties, A and B, now hold two positions (a long and a
short) with the exchange, which recognizes this and closes out A and B.
5.1.9 Consequences of failure to comply
We have described the various ways a trader would voluntarily close out
his position. A trader may also renege his obligation to replenish his ac-
count to its initial margin, when called by his broker to do so at some
time t
1
(t ` t
1
` T). Let us now consider what this trader's fate will be.
What would happen to a trader who suered losses, who saw his margin
account depleted beyond the maintenance margin, and who then refused
to post up the additional amount required at some time t
1
to bring his
account back to the initial margin level? Do not forget that at time t
1
, the
amount lost per unit by the (long) trader, for instance, is the dierence
between the initial futures price, F(tY T), and the futures price F(t
1
Y T); he
had committed himself to buying one unit of the commodity or the
nancial product at price F(tY T), and now (at time t
1
), the futures price
has fallen to F(t
1
Y T) ` F(tY T). He has therefore lost F(tY T) F(t
1
Y T),
and that amount has been withdrawn by the broker from his initial
deposit. The broker has the right to (and will) sell the client's contract at
the prevailing price, which is F(t
1
Y T). What will happen next?
Suppose that A is the defaulting party. Initially, his commitment to buy
was matched by the exchange against B's willingness to sell. The broker
has the power to transfer A's long contract to any other party (call this
other party C) at the prevailing price F(t
1
Y T) ` F(tY T). Party C now
becomes a long party, committing himself to buying the commodity or
the nancial product at the market futures price, F(t
1
Y T). B's position
was short and was matched initially to A's. Now C replaces A, so B and
C are matched together. The broker will refund to A the balance of his
account (on which A has lost F(tY T) F(t
1
Y T)B's gain), less a sizable
commission.
5.2 Pricing futures through arbitrage
An important relationship exists between a futures price and a spot price.
After all, for a perfectly dened product, the spot price paid today for said
commodity obviously has a strong relationship to the price set today for
the same commodity for a transaction that will take place in, say, six
months, since the only dierence in the contract is the time of delivery.
We will rst explain intuitively the kind of arbitrage that could take place
if a large dierence were to be observed between the futures and the spot
price. We will then be much more precise and evaluate exactly what that
dierence between the futures and the spot price, called the basis, should
be in equilibrium.
5.2.1 Convergence of the basis to zero at maturity
Let us rst observe that at maturity T the futures price F(TY T) must be
equal to the spot rate S(T). Indeed, any dierence would immediately
trigger arbitrage, which would force both prices toward the same level.
Consider the following example. Suppose that at maturity the futures
price is higher than the spot price, F(TY T) b S(T). We will show that
arbitrageurs would immediately step in, buying the commodity on the
spot market, selling it on the futures market, and pocketing the dierence.
They could do so through pure arbitrage, that is, without committing to
this operation any money of their own; they would simply borrow S(T)
for a day, buy the commodity, and sell the futures. At delivery, they
would cash in F(T) and pay back S(T) plus interesta very small inter-
est. Of course both supply and demand on each market would move in
such a way that these adjustments would be nearly instantaneous. On the
futures market, demand would shrink and supply would expand, thus
causing the futures price to fall, while on the spot market the contrary
would be truesupply would decrease and demand would increase,
entailing an increase in the spot price.
5.2.2 Properties of the basis before maturity, F(t, T) CS(t): introduction
Let us now consider that we are at time t, before T (T t = 6 months, for
instance). Suppose you observe a very large positive basisa huge dif-
ference between the futures price and the spot price. You could well ex-
ploit this dierence by buying the commodity on the spot market and
selling the futures.
To understand arbitrage on a futures contract, start with arbitrage on
the simpler forward contract. Suppose that the forward price is much
higher than the spot price. An arbitrageur would step in, sell the forward
contract, and buy the commodity on the spot market. At the maturity of
the contract, he would make delivery and pocket the dierence between
the forward price and the spot price. This would be his gross prot, from
112 Chapter 5
which a number of costs should be deducted: the interest he has to pay on
the loan he contracted to buy the commodity on the spot market, and
perhaps some storage, insurance, and transport costs. All these costs are
referred to as carrying costs, because they pertain to holding the com-
modity during the whole period (T t). Notice that if our arbitrageur did
not undertake a ``pure'' arbitrage, by borrowing p(tY T), but, on the con-
trary, used his own funds, he would still bear an interest cost, which we
would call an opportunity cost, because by investing his money in this
operation he loses the opportunity to invest it in some other vehicle. We
supposed that the forward price was way above the spot price, and
therefore we could well imagine that his gross prot would still be above
those costs. We could also suppose that the underlying asset is not a
commodity, but an income generating asset such as a stock (which might
bring a dividend during the holding period) or a bond (which might pay a
coupon). In such cases, the arbitrageur's costs would be reduced by such
revenues, and we can say that arbitrage will occur whenever the forward
price exceeds the spot price by more than these net costs (carrying costs
less possible income receipts from holding the asset).
We can now leave the case of arbitrage in the forward market to con-
sider the case of arbitrage on the dierence between a futures price and
the spot price of an asset. Let us not forget that since futures contracts are
marked to the market, the arbitrageur who has sold a futures contract and
who intends to deliver will not receive the initial futures price at delivery
time, but the futures price at maturity, that is, the spot price at maturity.
Indeed, any gain or loss at maturity will already have been settled through
the daily cash ows he has received or paid out; what counts at maturity
is the futures price at T, which equals the spot price, F(TY T) = S(T). We
will have the same outcome as in the case of a forward contract, but the
calculation of the arbitrageur's prot is somewhat dierent and goes
through the daily resettlements of the contract. Here is the process by
which an arbitrage on a forward contract and on a futures contract will
have the same outcome, if we do not take into account the possible in-
terest income ows generated by the daily settlements of the futures
contract.
When the futures contract matures, the short collects the net prots he
made on his futures contract: these are the net cash ows on his margin
account. As we have seen before, they add up to F(tY T) F(TY T) =
F(tY T) S(T). This amount is positive if the futures price contracted at
time t is larger than the futures price at T, equal to S(T); the net cash
ows are negative if F(tY T) ` S(T). Thus, the arbitrageur may very well
make a prot on the futures contract if F(tY T) b S(T), or suer a loss if
S(T) b F(tY T). However, since in this case he has bought initially, at
time t, the underlying commodity at S(t), which supposedly was much
lower than F(tY T), he will certainly make a prot.
Let us see why. At maturity T, his payo is made with two positions:
rst, his futures position, F(tY T) S(T), and second, his ``cash'' position.
Choosing to deliver, he will receive F(TY T) = S(T) and will remit the
commodity he has bought at price S(t); his cash position is S(T) S(t).
His payo is thus F(tY T) S(T) S(T) S(t) = F(tY T) S(t), a dif-
ference that corresponds to what he would have received in a forward
market and that is bound to be larger than the costs the arbitrageur will
incur. These costs are exactly the same as those that would prevail in an
arbitrage on a forward contract: carrying costs, less any income ow from
the stored asset. Since we have supposed that the dierence F(tY T) S(t)
was very large, it will admittedly be larger than those net costs. What will
be the outcome of such arbitrage? It will displace to the right the demand
curve on the spot market and also move to the right the supply curve on
Spot price
Spot market
Initial
basis
Equilibrium
spot price
Initial
(disequilibrium)
spot price
New
supply
Initial
supply
Initial demand
New demand
Initial
futures
price
S(t)
Futures price
Futures market
Equilibrium
futures
price
Equilibrium
basis
F(t,T)
Initial
supply
New
supply
Initial
demand
New
demand
Figure 5.3
Adjustment of the basis toward its equilibrium value. In the case of an exceedingly high initial
basis, supply and demand will shift both on the spot and futures markets. As a result, basis will
shrink to its equilibrium, equal to the cost of carry net of any possible income ows generated
by the asset.
114 Chapter 5
the futures market. This will tend to raise the price on the spot market
and lower it on the futures market, thereby decreasing the basis.
However, the basis will not necessarily wait for such an arbitrage to
shrink. Other forces will be at play. First, why should anybody buy the
commodity at such a highly priced futures or become a supplier in such a
bad spot market? Indeed, what will happen is that the demand will shrink
to the left on the futures market, because those who intended to hold the
commodity for six months will now consider buying it on the spot market,
and those who considered selling it now will try to postpone the sale. This
will have the same eect as the arbitrage we described before. Also, other
operators on both markets will reinforce the possible action of the arbi-
trageurs. Holders of the commodity will increase their supply on the
futures market at the expense of the supply on the spot market. So, arbi-
trageurs may well step into both markets (spot and futures) and lower the
basis by these actions, while the operators on any of those markets may
also contribute to the same kind of change.
The same kind of reasoning would apply if the basis were initially
extremely small or negative. Suppose, for instance, that it is zero
(F(tY T) = S(t)). What would be the outcome of such a situation? Reverse
processes such as the following could well take place. Those who own the
commodity could sell it and invest the proceeds at the risk-free rate. On
the other hand, they would buy a futures contract. At maturity time, they
would cash in their investment plus interest; they would then be able to
buy back their commodity at exactly the price of their investment and
keep the interest as net benet. This would tend to move to the right both
the supply curve on the spot market and the demand curve on the futures
market, thus exerting downward pressure on the spot price and upward
pressure on the futures price.
This is not, of course, the whole story. Arbitrageurs may very well step
in and borrow the commodity from owners of the good, short sell it, and
with the proceeds do what we have just described. After receiving the
commodity at the maturity of the futures contract, they would give back
the commodity to the owners. This supposes of course that it is indeed
possible to ``short sale'' the commodity (``short selling'' means borrowing
an asset and selling it, with the promise to restitute it at some later day).
Of course, normal transaction costs will be incurred by the arbitrageur,
and, as before, we should take into consideration any possible income
ow generated by the asset.
5.2.3 A simple example of arbitrage when the asset has no carry cost
other than the interest
In frictionless markets such as these, let us rst suppose that there are
no storage costs on commodities and that nancial assets that would be
traded on such spot and futures markets do not pay, within the time
period considered, any dividend (if these assets are stocks), nor do they
promise to pay any coupon (if the assets are bonds). We can now easily
see the theoretical relationship that will exist between the spot price and
the futures price. Let us go back to the situation where the futures price is
signicantly higher than the spot price multiplied by 1 i(0Y T)T.
By entering the series of transactions we described earlier, which we
summarize in table 5.2, we observe that with no negative cash ow at
time 0, an arbitrageur could generate a positive cash ow F(tY T) S(t)
(1 i(0Y T)T), since we supposed that F(tY T) b S(t) (1 i(0Y T)T).
This cannot last, since this arbitrageur would be joined by many others,
pushing up S(0) and driving down F(0Y T) until there could be no such
free lunch anymore; we would then have the basic relationship between
F(0Y T) and S(0):
F(0Y T) = S(0) (1 i(0Y T)T) (1)
Such a relationship could have been obtained in a straightforward way,
just by being careful with the units of measurement. Indeed, F(0Y T) is the
price of the commodity in terms of dollars to be paid at time T. S(0) is the
price of the same commodity at time t. Thus, the futures price can be none
other than the spot price, where dollars at 0 are converted into dollars at
Table 5.2
The futures buyer's positions: initial futures price: F(t,T)F$400//ounce on July 1
Time 0 Time T
Borrow S(0) at (simple)
interest rate i
0Y T
S(0)
Reimburse loan with
interest S(0) (1 i
0Y T
)T
Buy asset S(0) Sell asset at F(0Y T) F(0Y T)
Sell futures at F(0Y T) 0
Net cash ow 0 Net cash ow F
tY T
S
t
(1 i
0Y T
T)
116 Chapter 5
T. We must have a relationship of the form
$ at time T
commodity
=
$ at time 0
commodity

$ at time T
$ at time 0
The last expression on the right-hand side, ($ at time T)/($ at time 0), is
indeed a price: it is the price of $1 at time 0 expressed in dollars at T, and
this is equal to 1 i(0Y T)T.
Consider now the reverse situation, where the futures price F(0Y T) is
smaller than the future value of the spot price S(0)[1 i(0Y T)T[. An
arbitrageur could step in and proceed as follows. Let us suppose that if
he could dispose of all the proceeds of a short sale, he would borrow the
commodity (or the nancial asset), sell it on the spot market, and invest
the proceeds S(0) at rate i(0Y T) during period T. At the same time he
would perform this short sale, he would buy the commodity on the futures
market. At time T he would receive the result of his futures position,
which would be F(TY T) F(0Y T) = S(T) F(0Y T). He would also
collect the reimbursement of his loan plus interest, an amount equal to
S(0)[1 i(0Y T)T[; on the other hand, he would have to give back the
asset to the initial owner; he would have to buy the commodity on the
spot market for S(T) and give it back to its owner. In total his posi-
tion would be S(T) F(0Y T) S(0)(1 i(0Y T)T) S(T) = S(0)[1
i(0Y T)T[ F(0Y T). This operation is called reverse cash and carry.
We have supposed that S(0)[1 i(0Y T)T[ was larger than F(0Y T) and
have just shown that an arbitrageur could step in and pocket, as pure
arbitrage prot, that very dierence. Of course, all participants would
realize that excess supply would appear on the spot market, forcing its
price S(0) down; similarly, excess demand would build up pressure on the
futures market, driving its price F(0Y T) up until equilibrium, correspond-
ing to F(0Y T) = S(0)[1 i(0Y T)T[, would be restored.
Table 5.3 summarizes the initial and nal positions of an arbitrageur
who undertook these operations. As before, note that this is pure arbi-
trage, in the sense that no money outows are required from the arbi-
trageur at time 0, although a net prot is obtained at the futures contract
maturity.
5.2.4 Arbitrage with carry costs
In our introduction to futures pricing, we mentioned that the choice an
individual makes between taking positions either in the spot market or in
the futures market has to take into account the costs and the possible
benets of holding commodities or assets. Costs of holding assets always
include interest. According to the nature of assets (which may be nancial
assets or commodities ranging from precious metals to livestock), many
sorts of carry costs may be incurred, including storage costs, insurance
costs, and/or transportation costs. If the assets are of a nancial nature,
they may or may not carry a return. Finally, many commodities traded
on the futures markets enter a production process as intermediary goods
(e.g., gold, semiprecious metals, agricultural products). When held, these
goods generate costs and also a convenience return that is dicult to
measure. This convenience return corresponds to the benets any pro-
cessor of such goods receives by holding them in his inventories. Indeed,
let us suppose that the spot price of an intermediate good such as copper
is much too high compared to its futures price. Normally, what could
happen is this: pure arbitrageurs could step in, borrow copper from their
owners, short sell it, invest the proceeds at a risk-free interest rate, and
buy a futures contract; at maturity, they could receive (or pay) proceeds
of futures S(T) F(0Y T), collect proceeds of loan S(0)[1 i(0Y t)T[, and
buy asset [S(T)[, for a (positive) net cash ow of S(0)[1 i(0Y T)T[
F(0Y T). Also, owners of copper could themselves sell it on the spot
market, lend its proceeds, and buy a futures contract; at maturity, they
would receive the same cash ow as the pure arbitrageur. However, these
operations, which we may call pure arbitrage and quasi-arbitrage, respec-
tively, may not occur at all if owners of copper choose not to part with
Table 5.3
Transactions and corresponding cash ows at time 0 (today) and at time T in a simple reverse
cash-and-carry arbitrage
Time 0 Time T
Borrow asset
Sell asset
Lend proceeds
of asset's sale
Sell futures at
F(0Y T)
0
S(0)
S(0)
0
Receive proceeds of
futures short contract*
Collect loan plus
interest
Buy asset on the spot
market and remit it to
its owner
S(T) F(0Y T)
S(0)[1 i(0Y T)T[
S(T)
Net cash ow 0 Net cash ow S(0)[1 i(0Y T)T[ F(0Y t) b 0
*Sum of net cash ows received during time span (0Y T).
118 Chapter 5
their inventories, which may be trimmed down to their optimum level,
enabling their owners to meet any unforeseen excess demand in their
nal product, and thus preventing them from losing customers. There-
fore, inventories, which are assets, do bring a convenience income to their
owner. Although dicult to measure, we know that it exists, and it could
well prevent arbitrage or quasi-arbitrage in the event of a spot price that is
relatively high compared to a futures price.
In summary, we can classify carry costs and carry income as shown in
table 5.4, which indicates the kind of futures contract where each type of
cost or income typically applies.
Since in this book we deal mainly with futures on interest rates (on
bonds), we will not take into consideration this convenience income.
Thus, our main evaluation equation can be derived in a very natural way,
as follows. Remember that originally all futures markets were designed in
such a way as to enable future buyers and sellers of a commodity to hedge
against possible price uctuations, obviating all possible shortcomings of
a forward contract (i.e., a contract set on an unorganized market). Thus,
for any long party on the futures market, one unit of the commodity at
time T may be acquired in two ways. The rst is to take a position on the
futures market; the second is to acquire the commodity immediately (say
at time 0), incur carrying costs, and receive any carry income. The net
costs corresponding to both possibilities should be the same.
Let us take them in order. If our individual enters the futures market
with a long position, he will pay p(0Y T) S(T) at time T, which is
Table 5.4
Types of carry costs and carry income, with their corresponding futures contracts
Type of cost or income Type of futures contract where they typically apply
Carry costs
Interest All futures
Storage Commodities
Insurance Commodities
Transportation Commodities
Carry income
Dividend Stocks and stock indices
Coupons Bonds
Convenience income Commodities
the net cash ow of his futures position in future value p(0Y T) S(T),
the sum of positive and negative cash ows paid out during the futures'
contract life. We will assume that this is the future value of all these cash
ows. Of course, each cash ow should be evaluated in future value, since
the owner of the contract may incur positive and negative cash ows. The
error made by neglecting the interest corresponding to these cash ows
between their receipt and T is not very signicant. Furthermore, the long
party will buy the asset on the spot market and pay S(T). In all, he will
disburse p(0Y T) S(T) S(T) = p(0Y T).
Consider now the other way of acquiring the commodity at T and the
cost of that operation, evaluated at T. Our long party can buy the com-
modity outright on the spot market and pay S(0). The rst carry costs are
interest, equal to i(0Y T)S(0)T; these are evaluated with respect to time T
and are equal to the interest the long party has to pay out at T if he has
borrowed to pay the asset at price S(0). He will have opportunity costs as
well if he has used his own funds to buy the asset (he has had to forego
that interest). He may also incur minor carry costs such as the various fees
he will pay his broker, for instance, to keep the asset for him and collect
whatever income the asset can earn. We will call all those carrying costs in
future value CCFV (carrying costs in future value). On the other hand, if
the asset generates some income, in the form of dividends or coupons,
those payments have to be evaluated at time T; call them CIFV (carry
income in future value). The net cost in future value of buying the asset
outright on the spot market is therefore S(0) CCFV CIFV. Equilib-
rium requires that both prices be equal, and we should have a funda-
mental equilibrium relationship:
F(0Y T) = S(0) CCFV CIFV (2)
Note that this equation holds equally well for both the potential buyer
and the potential seller of the asset. Should there be any discrepancy be-
tween the left-hand side and the right-hand side of this equation, arbitrage
through the cash-and-carry mode or through reverse cash-and-carry can
be performed along the lines we described earlier, putting pressure on
both the spot price S(0) and the futures price F(0Y T), until equilibrium is
restored. We will explain next how arbitrageurs can operate in both cases.
120 Chapter 5
5.2.4.1 The case of an overvalued nancial futures contract
The case where the futures price is overvalued can be written as F(0Y T) b
S(0) CCFV CIFV. An arbitrageur would step in and take a short
position on the futures contract, selling the asset at time 0 at the futures
price F(0Y T). At the same time he would borrow S(0) in order to buy the
asset. None of these transactions would imply any net positive or negative
cash ow at time 0. At time T, our arbitrageur would liquidate his futures
short position; he would then receive a net cash ow equal to F(0Y T)
F(TY T) = F(0Y T) S(T). He would also sell the asset and receive S(T),
and then he would reimburse his loan of S(0), including the interest
S(0)i(0Y T)T plus fees to his broker (interest plus fees equal the carry
costs in future value, CCFV). Finally he would collect the income from
coupons or dividends received during the holding period of the asset,
which the arbitrageur invested until the maturity of the futures con-
tract. In all, these income ows are valued at CIFV at time T. (See table
5.5 for a summary of these transactions and their corresponding cash
ows.) Thus at time T he would receive F(0Y T) S(T) S(T) S(0)
CCFV CIFV = F(0Y T) S(0) CCFV CIFV, a positive amount
since we initially supposed that F(0Y T) b S(0) CCFV CIFV. Once
more we can see that an arbitrageur could turn any discrepancy between
Table 5.5
Transactions and corresponding cash ows in the case of a cash-and-carry arbitrage on an asset
with carry costs and carry income. Initial situation: futures price overvalued: F(0,T)I
S(0)BCCFVCCIFV
Time 0 Time T
Sell futures at
F(0Y T)
Borrow asset
value on spot
market
Buy asset on
spot market
0
S(0)
S(0)
Liquidate futures
short position
Sell asset on spot
market
Reimburse principal
on loan
Pay interest on
loan, plus fees to
stockbroker
Collect income
generated by asset
F(0Y T) S(T)
S(T)
S(0)
CCFV
CIFV
Net cash ow 0 Net cash ow F(0Y T) S(0) CCFV CIFV(b0)
two dierent prices for the same object into a net prot. Also we observe
that selling the asset on the futures market and buying it on the spot
market pulls down the futures price and pushes up the spot price of the
asset until equilibrium (as described by equation (1)) is restored.
Are there any uncertain elements in equation (1)? The only uncertainty
pertains to CIFV, the carry income in future value, since it does not rep-
resent coupons paid by default-free bonds, such as Treasury bonds. In-
deed, if the nancial instrument is a stock or an index on stocks, we
have to replace CIFV by its expected value, denoted by E(CIFV); prob-
ably the market will attach a risk premium (denoted ) in equation (1),
in the sense that this risk premium would have to be added to E(CIFV)
for (1) to hold. Thus, with uncertain cash income, we would have in
equilibrium:
F(0Y T) = S(0) CCFV E(CIFV) (3)
As always, the problem arises of evaluating this risk premium; matters
become complicated since transaction costs, and more generally carry
costs, are not the same for all actors in futures markets.
5.2.4.2 The case of an undervalued nancial futures contract
The case where the futures price is undervalued corresponds to F(0Y T) `
S(0) CCFV CIFV. In such circumstances arbitrageurs would buy the
futures for F(0Y T); they would then short sell the asset (they would bor-
row the asset, sell it on the spot market at S(0), and invest the proceeds).
At the maturity of the futures contract, arbitrageurs would settle their
long position, thus paying out F(0Y T) S(T) or receiving S(T)
F(0Y T). They would have to go on the spot market and buy the asset for
S(T), collecting interest on S(0) (presumably an amount very close to the
carry cost in future value that we mentioned earlier). The dierence be-
tween the borrowing rate and the lending rate times S(0) (the dierence
between the interest paid out by a short arbitrageur and the interest
received by a long arbitrageur) is more or less equal to the broker's fees
that the short arbitrageur had to pay. In other words, interest earned by
long arbitrageur = interest paid by short arbitrageur broker's fee =
CCFV. Finally, arbitrageurs would have to pay the asset's owner the
122 Chapter 5
carry income in future value generated during time span (0Y T) by the
asset borrowed at the outset, CIFV.
Table 5.6 summarizes these transactions and their outcome. At the
bottom line, the long arbitrageur would receive S(T) F(0Y T) S(T)
S(0) CCFV CIFV, equal to F(0Y T) S(0) CCFV CIFV, a
positive amount, since we have supposed that F(0Y T) ` S(0) CCFV
CIFV.
As before, the arbitrageur nets a prot equal to the dierence between
two values when they are out of equilibrium. Upward pressure on the
future price and downward pressure on the spot price will reestablish the
equilibrium relationship (1) between the two prices so long as there is no
uncertainty as to the future value of the carry income generated by the
asset.
5.2.5 Possible errors in reasoning on futures and spot prices
If the reader hesitates in his understanding of the various relationships
between futures prices, spot prices, and future spot prices, he may nd
solace in the fact that one of the most distinguished economists of the
twentieth century, John Maynard Keynes, erred on the subject; what's
more, it took seven decades for researchers to discover the error. Keynes'
Table 5.6
Transactions and corresponding cash ows in the case of a reverse cash-and-carry arbitrage on
an asset with carry costs and carry income. Initial situation: futures price undervalued:
F(0,T)HS(0)BCCFVCCIFV
Time 0 Time T
Buy futures
at p(0Y T)
Borrow asset
and sell asset
at S(0)
Lend asset's
value S(0)
0
S(0)
S(0)
Liquidate futures
long position
Buy asset on spot
market and return
it to its owner
Collect principal
and interest on
loan
Pay to the owner
of the asset income
generated by the
asset
F(0Y T) S(T)
S(T)
S(0) CCFV
CIFV
Net cash ow 0 Net cash ow F(0Y T) S(0) CCFV CIFV b 0
blunder on the subject was made in his Treatise on Money (1930); to
the best of our knowledge, this error was reported for the rst time by
Darrel Due in his remarkable book Futures Markets (1989), where it is
very well documented.
In his Treatise on Money, Keynes addresses the issue of the sign of the
basis (the dierence between the forward price and the spot price,
F(0Y T) S(0)). He denes ``backwardation'' as a situation in which the
spot price exceeds the forward price; in modern language, ``back-
wardation'' means that the basis, as we have dened it, is negative. For
Keynes, this is a normal situation, which he justies as follows:
It is not necessary that there should be an abnormal shortage of supply in order
that a backwardation be established. If supply and demand are balanced, the spot
price must exceed the forward price by the amount which the producer is ready to
sacrice in order to ``hedge'' himself, i.e., to avoid the risk of price uctuations
during his production period. Thus in normal conditions the spot price exceeds the
forward price, i.e., there is backwardation.4
There are some severe aws in this reasoning. First, there is no such risk
premium for the producer to sacrice in order to hedge himself; forward
and then futures markets have been developed precisely to enable both
the buyer and the seller of a commodity to protect themselves against
uncertainty without any cost. As we have seen, both agents have to ``pay''
something when they enter a forward or a futures market; they have to
forego possible gains at maturity T of the contract, but that is the dier-
ence between S(T) and F(tY T) (if it turns out to be positive) for the seller,
and the dierence between F(tY T) and S(T) (if it is positive) for the buyer.
However, this has nothing to do with the basis at t, F(tY T) S(t), which
may very well be positive or negative.
Furthermore, we can see easily that the only way to determine the sign
of the basis is writing equation (1) the following way:
F(tY T) S(t) = CCFV CIFV
In other words, in equilibrium the sign of the basis is just the sign of the
dierence between carry costs and carry income in future value. When no
4 John Maynard Keynes, A Treatise of Money, Vol. II, chapter 2a, ``The Applied Theory of
Money,'' part V, ``The Theory of the Forward Market''; quoted by Darrell Due in his
Futures Markets (Englewood Clis, N.J.: Prentice Hall, 1989), p. 99.
124 Chapter 5
seasonal shortage or excess supply exists or is predictable, there are many
commodities for which carry costs are denitely larger than carry income,
and this is why under normal conditions, contrary to what Keynes ascer-
tained, the normal situation is that of a positive basis.
5.3 The relative bias in the T-bond futures contract
5.3.1 The various types of options imbedded in the T-bond futures contract
We should underline that the short trader, who has committed himself to
selling an eligible T-bond (with at least 15 years maturity or 15 years to
rst callable date) on a delivery day, holds a number of valuable options.
First, he has what is called a wild card option. This in turn can be broken
up into two options: the rst is the time to deliver option; the second is the
choice of the bond to be delivered (also called a quality option).
5.3.1.1 The wild card (time to deliver) option
The delivery of the T-bond by the short party must be done in any busi-
ness day of the delivery month. However, the various operations needed
for that span a three-day period:
.
The rst day is called position day; it is the rst permissible day for the
trader to declare his intent to deliver. The rst possible position day is the
penultimate business day preceding the delivery month. The importance
of this ``position day'' is considerable: on that day at 2 p.m. the settlement
price and therefore the invoice amount to be received by the short are
determined (hence the day's name). Now the short can wait until 8 p.m. to
announce her intent to deliver. Clearly, during the six hours, she has the
option of doing precisely this, thereby reaping some prots (or not). If
interest rates rise between 2 p.m. and 8 p.m., the bond's price will fall. She
will then earn a prot from the dierence between the actual price she will
receive for the bond (called the settlement price), set at 2 p.m., and the
prices he will have to pay to acquire the bond, which will be lower than
the settlement price because of the possible rise in interest rates between
2 p.m. and 8 p.m.
.
The second day has only administrative importance; it is the notice of
intention day. During this day the clearing corporation identies the short
and the long trader to each other.
.
The third day is the delivery day. During this day, the short party is
obligated to deliver the T-bond to the long party, who will pay the set-
tlement price xed at 2 p.m. on position day.
The short trader has the option to exercise this ``time to deliver'' option
from the penultimate business day before the start of the delivery month
and the eighth to the last business day of the delivery month, called the
last trading day. On that day at 2 p.m. the last settlement price is estab-
lished for the remainder of the month. This still leaves our short with a
special option, called an end-of-the-month option, which we examine now.
Suppose our short has not elected to declare her intention to deliver
until the last trading day, either because interest rates have not increased
or because they have not increased enough to her taste. The last trading
day has now come, and the settlement price she will receive when deliver-
ing the bond is now xed. She can still wait ve days to declare her in-
tention. This leaves her with the possibility of earning some prots if, for
instance, she buys the bond at the last trading day settlement price, keeps
it for a few days, and therefore earns accrued interest on it. She will re-
ceive a prot if the accrued interest is higher than the interest she will pay
in order to nance the purchase of the bond.
The various options oered to the short party carry value; this value
added to the futures price must equal the present value of the bond plus
carry costs in order to yield equilibrium values both for the bond and
the futures. In the last 20 years, many studies have been undertaken to
capture this rather elusive option value. Robert Kolb made an excellent
synthesis of these researches in his text Understanding Futures Markets.5
We will now take up the issue of analyzing in depth the relative bias
introduced by the Chicago Board of Trade in the T-bond futures conver-
sion factor.
5.3.1.2 The quality option
A short party has the benet of exercising another important option: he
can choose to deliver a T-bond among a list of bonds, established regu-
larly by the Chicago Board of Trade, which have at least 15 years of
maturity from their delivery date until their rst call date. Since the bonds
can vary widely in maturity and coupon rate, the CBOT normalizes their
5 R. Kolb, Understanding Futures Markets (Kolb Publishing Company, 1994).
126 Chapter 5
value by applying to each a conversion factor (described in detail in the
appendix to this chapter).
It is well known that when yields are above 8%, the Chicago Board of
Trade conversion factor introduces a bias that tends to favor the delivery
of low coupon, long maturity bonds; the opposite is true when yields are
below 8%. In this section we will show why this is so. The main reason for
this is that any conversion factor system (as set up by the Chicago Board
of Trade, or any other exchange organizing a futures market on long-term
government securities along the same lines) introduces a relative bias that
is an ane transformation of the duration of the bond to be remitted. The
transformation is a positive one when rates are above 8% and a negative
one when rates are below 8%.
Since the introduction of the CBOT T-bond futures contract, excellent
studies of the bias entailed by the conversion factor have been carried out
(see in particular M. Arak, L. S. Goodman, and S. Ross (1986); T. Kill-
collin (1982); J. Meisner and J. Labuszewski (1984)).6 However, these
researchers concentrated on the simple dierence between the CBOT
conversion factor and the ``fair,'' or theoretical, factor. It is of crucial
importance to consider, in fact, the relative dierence between the CBOT
factor and the fair factor, because this relative bias can be shown to be an
ane transformation of the remitted bond's durationgiving the key to
prot maximization in the delivery process, at least in an environment of
at term structures.
5.3.2 The conversion factor as a price
The conversion factor calculated by the CBOT is basically the value of a
$1 principal bond, carrying a coupon c (as a percentage) and a maturity
T equal to those of the bond remitted by the seller, when the yield to
maturity is 6%.7 In order to understand the economic meaning and the
nancial implications of converting the futures price into an invoice
amount the long party will have to pay, the following notation is useful:
6 Marcelle Arak, Laurie S. Goodman, and Susan Ross, ``The Cheapest to Deliver Bond on
the Treasury Bond Futures Contract,'' Advances in Futures and Options Research, Volume 1,
Part B (Greenwich, Conn.: JAI Press, 1986), pp. 4974; Thomas E. Killcollin, ``Dierence
Systems in Financial Futures Contracts,'' Journal of Finance (December 1982): 11831197;
James F. Meisner and John W. Labuszewski, ``Treasury Bond Futures Delivery Bias,''
Journal of Futures Markets (Winter 1984): 569572.
7 The Chicago Board of Trade began calculating the conversion factor this way on March 1,
2000.
B
cY T
(i) = value of a $100 principal bond, with coupon c (in %) and
maturity T, when the structure of interest rates is at and equal to i;
coupon is paid semi-annually.
For instance, B
9%Y 20
(6%) is the value of a bond with 9% semi-annual
coupon, 20-year maturity, when the rate of interest is 6%. In such a case,
B
9%Y 20
(6%) is 137.90. Using this notation, we can write the conversion
factor CF
cY T
as
Conversion f actor ICF
cY T
=
B
cY T
(6%)
100
(Note that the conversion factor depends solely on the coupon and ma-
turity of the delivered bond and not on the rate of interest; we denote it
CF
cY T
.) In the case of the bond we just considered, the conversion factor
would be 1.3790. For bonds with a maturity not equal to an integer
number of years, the conversion factor can be determined from Appendix
5.A.1, a portion of the conversion table of the Chicago Board of Trade.
The conversion factor can thus be seen as the ratio of two T-maturity
bond prices: the price of a c-coupon evaluated at 6% and the price of a
6%-coupon evaluated at 6% (which is equal to 100). The ratio of these
two prices is therefore an exchange rate between a (6%Y 20) bond, and a
(cY T) bond when i is 6%. We have
CF =
B
cY T
(6%)
B
6%Y 20
(6%)
=
$
(cY T) bond
$
(6%Y 20) bond
=
number of (6%Y 20) bonds
one (cY T) bond
Being an exchange rate, the conversion factor is therefore a price: the
price of one (cY T) bond, expressed in 6%-coupon bonds when rates are
equal to 6%. If one remits a (cY T), one must count it as if it were a
number CF of (6%Y 20) bonds, which themselves are evaluated at the
futures prices.
5.3.3 The ``true'' or ``theoretical'' conversion factor
The above-mentioned property of the conversion factor suggests that the
conversion factor may well introduce a bias as soon as the rate of interest
is not 6%. Consider, for instance, the case where the interest rate structure
is at 4%, and where the seller chooses to remit a (10%Y 20) bond. The
conversion factor is entirely independent from the interest rate structure;
it is xed at CF
cY T
=
B
10%Y 20
(6%)
100
= 1X4623. However, let us now introduce
128 Chapter 5
the concept of a ``true'' (or theoretical) conversion factor, denoted as
TCF
cY T
(i), the number of (6%Y 20) bonds that correspond to one (10%Y T)
bond. Such a true conversion factor should be dened as follows:
TCF
cY T
(i) =
number of (6%Y 20) bonds at 4%
one (10%Y 20) bond at 4%
=
price of a (10%Y 20) bond at i = 4%
price of a (6%Y 20) bond at i = 4%
=
B
10%Y 20
(4%)
B
6%Y 20
(4%)
=
182X07
127X36
= 1X4296
for a relative dierence (CF TCF)aTCF of 2.28%.
More generally, the theoretical conversion factor for delivering a
(c%Y T) bond when the interest rate is i should be equal to the number of
(6%Y 20) bonds that a (c%Y T) bond is worth. Hence it is equal to the price
of a (c%Y T) bond divided by the price of a (6%Y 20) bond when rates are
i. We should have
TCF
cY T
(i) =
number of (6%Y 20) bonds
one (c%Y T) bond
=
price of (c%Y T) bond at rate i
price of (6%Y 20) bond at rate i
=
B
c%Y T
(i)
B
6%Y 20
(i)
5.3.4 The relative bias in the conversion factor
We now introduce the concept of the relative bias in the CBOT conver-
sion factor, dened as the dierence between the conversion factor and
the true conversion factor, divided by the true conversion factor:
Relative bias Ib =
CF TCF
TCF
=
CF
TCF
1 I
B
c%Y T
(6%)
B
6%Y 20
(6%)
B
c%Y T
(i)
B
6%Y 20
(i)
1 (4)
Of course, there is no bias if i = 6%, because then TCF = CF. How-
ever, this is not the only case where the conversion factor does not intro-
duce a bias, although this is seldom realized. We will show that there is
in fact a whole set of values (rate of interest i, coupon c, and maturity of
the deliverable bond T) that lead to a zero relative bias; these are all the
values of i, c, and T for which the equation b = 0 is satised. To get a
clear picture of this, it is useful rst to determine, for any given value of i,
what are the limits of the relative bias introduced by the conversion factor
when the maturity of the bond tends toward either innity or zero. Al-
though deliverable bonds must have a maturity longer than 15 years and
are usually shorter than 30 years, we gain in our understanding of the
relative bias by considering a broader set of bonds and then retaining only
those of interest.
Consider rst what would be the bias in the conversion factor when
the maturity of the deliverable bond goes to zero. Since lim
T0
B
c%Y T
(6%) =
lim
T0
B
c%Y T
(i) = 100, we can write, from (4),
lim
T0
b =
B
6%Y 20
(i)
100
1
For example, if i = 3%, the relative bias tends toward 44X9%; if
i = 9%, the bias is 27X6%. This limit is positive if and only if i ` 6% and
negative if and only if i b 6%.
We can see that this limit depends solely on i and not on the coupon c.
It implies that all curves representing the bias for dierent coupons will
intersect on the ordinate (for T = 0). This is shown in gures 5.4 and 5.5,
which correspond to possible biases for various coupons and maturities,
and for interest rates equal to i = 9% and i = 3%, respectively.
Let us now see what the relative bias is if we deliver a bond with innite
maturity. Since lim
Ty
[B
c%Y T
(6%)aB
c%Y T
(i)[ =
c
6%
_
c
i
=
i
6%
, we have
lim
Ty
b =
i
6%
B
6%Y 20
(i)
100
1
Denoting 1a(1 ia2)
40
as 0, we have, after simplications, lim
Ty
b =
0(
i
6%
1). Thus lim
Ty
b is positive if and only if i b 6%, and negative if
and only if i ` 6%.
For instance, if i = 3%, the limit of the bias is 27X6%; if i = 9%, it is
equal to 8.6%. This limit does not depend on the coupon rate either. It
implies that the relative bias of each bond in (maturity, bias) space will be
represented by a series of curves that have nodes (intersections) at the
origin and at innity. It is important to realize that, for i b 6%, the ordi-
130 Chapter 5
nates of these nodes are respectively negative (for T 0) and positive (for
T y). The function b being continuous in T, we can apply Rolle's
theorem and conclude that b will at least once be equal to zero (its curve
will cross the abscissa at least once). This is conrmed by gure 5.4: for
any coupon c, and for i = 9%, the relative bias is zero for one value of T,
which depends upon c. The same is true when i ` 6%: the ordinates of the
nodes are this time respectively positive (when T 0) and negative (when
T y). Therefore, b will cross the abscissa at least once, as is conrmed
in gure 5.5, corresponding to i = 3%. For each deliverable bond, there
will be one maturity for which the CBOT conversion factor does not
entail any bias, even if i is dierent from 6%.
Figure 5.4 represents relative biases for bonds with semi-annual cou-
pons of 2, 4, 6, 8, and 12%, when the rate of interest is equal to 9%. For
all bonds with a coupon lower than the rate of interest (for all below-par
bonds), the relative bias always rst increases, goes through a maximum,
and then tends toward the common limit mentioned above. For all bonds
0 20 40 60 80 100
20
10
0
10
20
30
Maturity of delivered bond (years)
R
e
l
a
t
i
v
e

b
i
a
s

(
i
n

%
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 5.4
Relative bias of conversion factor in T-bond contract, with i F9%.
above par (in this case, the 8% and 12% coupon bonds), the movement
toward the limit is monotonous; the bias constantly increases along a
curve that is attening out.
Do these curves on gure 5.4 look familiar? They may, because this
kind of behavior is fundamentally that of a duration. Indeed, we will now
show that the relative bias of the conversion factor in the Chicago Board
of Trade T-bond contract (or any contract of that type) is an ane
transformation of the duration of the deliverable bond; it is a negative
transformation when i ` 6%, and it is a positive one when i b 6%. (From
a geometric point of view, a negative ane transformation means that the
graph of the relative bias is essentially the graph of the duration of the
deliverable bond, upside down; the scaling is modied both linearly and
by an additive constant.)
First, let us denote the modied duration of a bond with value B
cY T
(i
0
),
as D
cY T
(i
0
). From the properties of modied duration, we know that it is
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
0 20 40 60 80 100
50
40
30
20
10
0
10
20
30
40
R
e
l
a
t
i
v
e

b
i
a
s

(
i
n

%
)
Figure 5.5
Relative bias of conversion factor in T-bond contract, with i F5%.
132 Chapter 5
equal, in linear approximation, to the relative change in the bond's value
per unit change in the rate of interest.
B
cY T
(i) B
cY T
(i
0
)
B
cY T
(i
0
)
e
dB
B
= D
cY T
(i
0
) (i i
0
) (5)
Therefore, changing i i
0
into i
0
i, the ratio of two values of the bond,
corresponding to two dierent rates of interest, can be written as
B
cY T
(i)
B
cY T
(i
0
)
e1 D
cY T
(i
0
) (i
0
i) (6)
Let us now come back to the CBOT conversion factor. Its relative bias
is equal to
b =
B
c%Y T
(6%)
B
6%Y 20
(6%)
B
c%Y T
(i)
B
6%Y 20
(i)
1 (7)
which can be written equivalently as
b =
B
cY T
(6%)
B
cY T
(i)
B
6%Y 20
(6%)
B
6%Y 20
(i)
1 e
1 D
cY T
(i) (i 6%)
1 D
6%Y 20
(i) (i 6%)
1 (8)
or, more compactly, as
b = mD
cY T
(i) n
where
mI
(i 6%)
1 D
6%Y 20
(i) (i 6%)
and
n I
1
1 D
6%Y 20
(i) (i 6%)
1
We can check that the bias is zero whenever i = 6%. Also, we can easily
show that for any relevant value of i, the denominator (m) cannot become
negative. Therefore, the relative bias essentially behaves like an ane
transformation of the duration of the deliverable bond: a positive one
(m b 0) whenever i b 6% and a negative one (m ` 0) when i ` 6%. This
explains the shapes of the bias that correspond to an interest rate either
above 6% (9%; gure 5.4) or below 6% (3%; gure 5.5).8
There are many lessons to be learned from these diagrams. Consider
rst the case i b 6%X Notice that the conversion factor unequivocally
favors bonds with high duration. This stems from the fact that, as we have
just shown, the bias in this case is an ane transformation of the duration
of the delivered bond. It is crucial to observe here that contrary to the
generally held view, this does not mean that the bias favors high maturities
bonds. Indeed, it is well known that for under-par bonds (bonds such that
their coupon is below the rate of interest), duration goes through a maxi-
mum (whose abscissa cannot be given in analytic form, but which may
very easily be computed). It then converges toward the common limit
equal to
1
i
:, where : is the fraction of a year remaining until payment
of next coupon. (Here : is
1
2
.) Consider, for example, a very low coupon
bond (2%), with i = 9%. The relative bias for such a bond will go through
a maximum for a maturity of 36 years (rounded to the full year), and we
can very well surmise that the prot of delivering that bond will also go
through a maximum. This is what we will show in the next section.
If we now turn to the case where i is below 6% (i = 3%; gure 5.5), the
opposite is true: the bias tends to favor high-coupon, low-duration bonds
because the bias behaves like the negative of a duration. This implies that,
in a symmetrical way to what we have just seen for i b 6%, low-coupon
bonds will see their bias go through a minimum before tending toward
their common limit. This implies that we will have to be careful about all
these factors when analyzing the various prots that will correspond to all
deliverable bonds.
5.3.5 The prot of delivering and its biases
A bias in the calculation of the amount to be paid by the long party in the
delivery process may well entail some bias in the prot earned by the
short party. We will rst recall what this prot is equal to. We will then
8 It would be possible to increase the exactness of the relationship between the relative bias
and the duration measure by adding (positive) convexity terms both in the numerator and in
the denominator of the fraction in the right-hand side of (8). The conclusions would of
course remain the same.
134 Chapter 5
show the considerable biases that are introduced by the conversion factor,
as well as the kind of surprises they carry with them.
The prot of the short is basically equal to the amount of the invoice
paid by the long to the short, less the cost of acquiring the deliverable
bond. Referring to a c-coupon, T-maturity deliverable bond, and again
keeping the hypothesis of a at structure rate i, the prot
cY T
(i) of
delivering the bond is a function of the three variables cY T, and i, equal to
cY T
(i) = F CF
cY T
B
cY T
(i) (9)
where F denotes the futures price at maturity.
It is useful to write out this function. Remember that F and CF are
given by
F = B
6%Y 20
(i)
and
CF =
B
cY T
(6%)
100
respectively.
Thus we have
cY T
(i) = B
6%Y 20
(i)
B
cY T
(6%)
100
B
cY T
(i) (10)
Replacing the three functions on the right-hand side of (10) by their
expressions, we have
cY T
(i) =
6
i
1
1
(1 ia2)
40
_ _
100
(1 ia2)
40
_ _
ca100
0X06
1
1
1X03
2T
_ _
1
1X03
2T
_ _
c
i
1
1
(1 ia2)
2T
_ _
100
(1 ia2)
2T
_ _
(11)
This function of i, c, and T has some important properties, which can best
be described through gures 5.6 and 5.7.
The rst property to notice is that there is a whole set of values (cY T)
that, for given i, lead to a zero prot. Those are the values we dened
before as entailing a zero relative bias in the conversion factor. Suppose,
indeed, that c and T are such that the relative bias is zero. This means
that the conversion factor is equal to the true conversion factor, which we
had found to be equal to the number of (6%Y 20) bonds exchangeable for
one (c%Y T) bond, that is, the ratio of the price of a (c%Y T) bond divided
by the price of a (6%Y 20), both evaluated at rate i (TCF = B
c%Y T
(i)a
B
6%Y 20
(i)). Let us now replace in the prot function [
cY T
(i) in equation
(9)] the conversion factor CF by the true conversion factor. We thus get a
prot equal to zero, as foreseen:
cY T
(i) = F(T) TCF B
cY T
(i)
= B
6%Y 20
(i)
B
cY T
(i)
B
6%Y 20
(i)
B
cY T
(i) = 0 (12)
0 20 40 60 80 100
20
10
0
10
20
30
P
r
o
f
i
t

(
$
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 5.6
Prot from delivery with T-bond futures; F(T)F75.93; i F11%. Note that the relative bias of
the conversion factor for a 6% coupon, 20-year maturity bond is zero, as it should be.
136 Chapter 5
The second important property to notice is that not all prot functions
are monotonously increasing or decreasing; this is partly due to the non-
monotonicity of the conversion factor, which, as we saw earlier, is basi-
cally an ane transformation of duration, itself a nonmonotonic function
for all bonds with nonzero coupons below the rate of interest.
Consider rst the case i = 11% (i b 6%). All prots for coupons equal
to or higher than 11% are always increasing (this is the case in gure 5.6
for c = 12). On the other hand, all prots for coupons below 11% are
increasing in a rst stage, going through a maximum, and then decreas-
ing. For any coupon, each prot tends toward its own limit, given by
lim
Ty
cY T
(i) = c
B
6%Y 20
(i)
6

1
i
_ _
(13)
0 20 40 60 80 100
50
0
50
100
150
P
r
o
f
i
t

(
$
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 5.7
Prot from delivery with T-bond futures; F(T)F137.65; i F5%. Note that, as in the preceding
case where i was equal to 9%, the relative bias of the conversion factor when remitting a 6%,
20-year bond is zero, as it should be.
which (contrary to the corresponding limit for the relative bias) depends
on the coupon. For example, if c = $12, this limit is $4.81.
We now turn toward the case i = 3% (i ` 6%). The prot functions are
always decreasing for coupons below 5%; for coupons above 5%, they rst
decrease, reach a minimum value, and then increase. As before, for any
coupon, each prot tends toward its own limit, given by (13).
It is crucial to recognize that all prot curves intersect at two points
(nodes). The rst node is xed at T = 0; the second one, for usual values
of i, is not very sensitive to i and lies between 31.226 years for i = 5% and
34.793 years for i = 11% (see gures 5.7 and 5.6 respectively).
The general shape of these functions and the second node in the prot
curves implies possible reversals in the ranking of the cheapest (or most
protable) to deliver bonds. Consider rst an environment of high interest
rates. Until now, as we noted in our introduction, it was believed that
in such a case the most protable bond to deliver was the one with the
lowest coupon and the longest maturity. This is clearly not always true.
Consider, for instance, a low coupon bond (c = 2%). The prot from
delivering it, if i = 11%, goes through a maximum when its maturity is (in
full years) 24 years and is equal to $3.2058. This is larger than the prot of
remitting the same coupon bond with a 30-year maturity, which would be
$2.9206 (a relative dierence of 9.8%). From what we said earlier, we
might guess that the 24-year bond has a higher duration than the 30-year
bond. (This is indeed true: D
2%Y 24
(11%) = 12X07 years, and D
2%Y 30
(11%)
= 11X75 years.) So if a rule of thumb is to be used, we should always rely
on duration, not on maturity.
The second observation we make is quite surprising and is due to the
existence and characteristics of the second node in the prot functions.
We can see that should the maturity of deliverable bonds be beyond the
second node, prot reversals would be the rule whatever the rates of interest
(with the exception, of course, of i = 6%). For any maturity beyond the
second node, the bond that is cheapest to deliver would be the one with
the highest coupon when i b 6% and the one with the lowest coupon
when i ` 6%. This property would become crucial if some foreign trea-
sury issued bonds with maturities exceeding 30 years and if those bonds
were deliverable in a futures market. This would be of particular impor-
tance if interest rates were above 6% (see gure 5.6, where a high coupon
bond's prots will dominate the prots that can be made from delivering
any other bond).
138 Chapter 5
Naturally, this analysis applies when we consider a at term structure.
For completeness, it should be extended to nonat structures. This would
be a formidable taskbut we would certainly not be surprised to discover
results similar to those obtained with at structures.
Appendix 5.A.1 The calculation of the conversion factor in practice
The conversion factor used by the Chicago Board of Trade is the theo-
retical value of a $1 nominal, c% coupon, T-maturity bond when its
yield to maturity is 6%. We will now show how this conversion factor is
calculated.
First, the remaining maturity in months above a given number of years
for the deliverable bond is rounded down to the nearest three-month
period. Two cases may arise:
1. The remaining maturity is an integer number of years T plus a number
of months strictly between 0 and 3, or between 6 and 9. In this case, the
retained maturity (in years) for the calculation will be T or T
1
2
, respec-
tively; the number of six-month periods will then be n = 2T or n =
2T 1, respectively. Formula (14) will then apply for a $1 bond maturing
in number n of six-month periods, which will be equal to either 2T or
2T 1. In this case the conversion factor (CF) will be
CF =
n
t=1
c
2
(1X03)
t
(1X03)
n
Y n = 2T or 2T 1 (14)
We then have
CF =
ca2
0X03
1
1
1X03
n
_ _
1X03
n
Y n = 2T or 2T 1 (15)
For example, if a 9% yearly coupon rate bond has a remaining maturity
equal to 24 years and 8 months, rounding down to the nearest three-
month period gives 24 years and 6 months, which is equivalent to 49
six-month periods. Hence n = 49 in formulas (14) and (15), and the con-
version factor is equal to CF = 1X3825, as can be veried from the table
published by the Chicago Board of Trade (see table 5.A.1).
2. The remaining maturity is an integer number of years T plus a number
of months between 3 and 6, or between 9 and 12. In such circumstances,
Table 5.A.1
Extract from the conversion table used by the Chicago Board of Trade for the Treasury Bond
Futures contract (as of March 1, 2000)
YearsMonths
Coupon 259 256 253 250 249 246 243 240
8 1.2604 1.2595 1.2583 1.2573 1.2560 1.2550 1.2537 1.2527
8
1
8
1.2767 1.2757 1.2744 1.2734 1.2720 1.2710 1.2696 1.2685
8
1
4
1.2930 1.2920 1.2906 1.2895 1.2880 1.2869 1.2854 1.2843
8
3
8
1.3093 1.3082 1.3067 1.3055 1.3040 1.3028 1.3013 1.3000
8
1
2
1.3256 1.3244 1.3229 1.3216 1.3200 1.3188 1.3172 1.3158
8
5
8
1.3419 1.3406 1.3390 1.3377 1.3361 1.3347 1.3330 1.3316
8
3
4
1.3582 1.3568 1.3552 1.3538 1.3521 1.3506 1.3489 1.3474
8
7
8
1.3744 1.3730 1.3713 1.3699 1.3681 1.3666 1.3647 1.3632
9 1.3907 1.3893 1.3875 1.3859 1.3841 1.3825 1.3806 1.3790
9
1
8
1.4070 1.4055 1.4036 1.4020 1.4001 1.3985 1.3965 1.3948
9
1
4
1.4233 1.4217 1.4198 1.4181 1.4161 1.4144 1.4123 1.4106
9
3
8
1.4396 1.4379 1.4359 1.4342 1.4321 1.4303 1.4282 1.4264
9
1
2
1.4559 1.4541 1.4520 1.4503 1.4481 1.4463 1.4441 1.4422
9
5
8
1.4722 1.4704 1.4682 1.4664 1.4641 1.4622 1.4599 1.4580
9
3
4
1.4884 1.4866 1.4843 1.4824 1.4801 1.4782 1.4758 1.4738
9
7
8
1.5047 1.5028 1.5005 1.4985 1.4961 1.4941 1.4917 1.4895
10 1.5210 1.5190 1.5166 1.5146 1.5121 1.5100 1.5075 1.5053
10
1
8
1.5373 1.5352 1.5328 1.5307 1.5282 1.5260 1.5234 1.5211
10
1
4
1.5536 1.5515 1.5489 1.5468 1.5442 1.5419 1.5392 1.5369
10
3
8
1.5699 1.5677 1.5651 1.5628 1.5602 1.5578 1.5551 1.5527
10
1
2
1.5861 1.5839 1.5812 1.5789 1.5762 1.5738 1.5710 1.5685
10
5
8
1.6024 1.6001 1.5974 1.5950 1.5922 1.5897 1.5868 1.5843
10
3
4
1.6187 1.6163 1.6135 1.6111 1.6082 1.6057 1.6027 1.6001
10
7
8
1.6350 1.6326 1.6297 1.6272 1.6242 1.6216 1.6186 1.6159
11 1.6513 1.6488 1.6458 1.6432 1.6402 1.6375 1.6344 1.6317
Source: Chicago Board of Trade. Website: http://www.cbot.com/ourproducts/nancial/
US6pc.html
140 Chapter 5
the Chicago Board of Trade considers that the maturity (in years) is either
T plus three months, or T
1
2
plus three months; hence the bond is
considered to have n = 2T six-month periods plus three months, or n =
2T 1 six-month periods plus three months respectively. The calculation
for the $1 face value bond then amounts to performing the following four
evaluating operations:
.
evaluate the bond at a three-month horizon from now;
.
add the six-month coupon to be received at that time;
.
discount that sum to its present value;
.
subtract (pay o) half a six-month coupon, considered as accrued in-
terest on the bond (since it does not belong to the bond's owner).
These four operations are summarized in the following formula:
CF =
ca2
0X03
(1
1
1X03
n
) 1X03
n
c
2
1X03
1a2

c
4
(16)
As an example, consider a 9% yearly coupon rate bond with an
(observed) remaining maturity of 24 years and ve months, which is then
rounded down to 24 years and three months. Hence, the number of six-
month periods is 48, and the retained maturity is 48 periods plus three
months. Formula (16) applies, yielding a conversion factor CF = 1X3806,
as shown in table 5.A.1.
Main formulas
.
Futures price:
F(0Y T) = S(0) CCFC CIFV
where
CCFV Icarry costs in future value
CIFV Icarry income in f uture value
.
Conversion factor:
CF
cY T
=
B
cY T
(6%)
100
.
``True'' or ``theoretical'' conversion price:
TCF
cY T
=
B
cY T
(i)
B
6%Y 20
(i)
.
Relative bias in the conversion factor:
b =
CF TCF
TCF
=
B
c%Y T
(6%)
B
6%Y 20
(6%)
B
c%Y T
(i)
B
6%Y 20
(i)
1
emD
cY T
(i) n
where
m =
i 6%
1 D
6%Y 20
(i) (i 6%)
and
n =
1
1 D
6%Y 20
(i) (i 6%)
1
Questions
The following exercises should be considered as projects, to be done
perhaps with a spreadsheet.
5.1. Determine the relative bias in the conversion factor and the prot
from delivering T-bonds with coupons 2%, 4%, 6%, 8%, and 12% when
the rate of interest is at 10%. Draw a chart of your results, similar to g-
ures 5.4 and 5.6.
5.2. Determine the relative bias in the conversion and the prot of deliv-
ering the same bonds as in (5.1), when the interest rate is 5%. Draw charts
of your results, similar to gures 5.5 and 5.7.
142 Chapter 5
6
IMMUNIZATION: A FIRST APPROACH
Alonso. Irreparable is the loss; and patience
Says it is past her cure.
The Tempest
Ghost of Lady Anne. Thou quiet soul, sleep thou a quiet sleep
Dream of success and happy victory!
Richard III
Chapter outline
6.1 An intuitive approach 145
6.2 A rst proof of the immunization theorem 148
6.2.1 A rst-order condition 149
6.2.2 A second-order condition 150
Overview
Holding bonds may be very rewarding in the short run if interest rates go
down; but if they go up, the owner of a bond portfolio may suer sub-
stantial losses. This is the short-run picture. However, the opposite con-
clusions apply if we consider long horizons: a decrease in interest rates
results in a decrease in the total return on a bond, because the value of the
bond, close to maturity, will have come back close to its par value, and the
coupons will have been reinvested at a lower rate. On the other hand, if
interest rates go up, the value of the bond may not be aected in the long
run (because its value will have gone back to its par value), but the coupons
will have been reinvested at a higher rate. A natural question then arises: is
there an `ìntermediate horizon'' for any bond, between the ``short term''
and the ``long term,'' such that any move of the interest rate immediately
after the purchase of the bond will be of no consequence at all for the
investor's performance? This intermediate horizon does exist, and we will
show in this chapter that it is equal to the duration of the bond.
If an investor has a horizon equal to the duration of the bond, his in-
vestment will be protected (or `ìmmunized'') against parallel shifts in the
structure of interest rates. Of course, the second question that comes to
mind is: what if the investor's horizon diers from the duration of the
bonds he is willing to buy because they seem favorably valued and they
are reasonably liquid? The answer is to construct a portfolio of bonds
such that the duration of the portfolio will be equal to the investor's
horizon. The duration of the portfolio is simply the weighted average of
the durations of the individual bonds, and the weights are the proportions
of each bond in the portfolio.
Immunization is then dened as the set of bond management proce-
dures that aim at protecting the investor against changes in interest rates.
They amount essentially to making the duration of the investor's portfolio
equal to his horizon. Immunization is denitely a dynamic set of proce-
dures, because the passage of time and changes in interest rates will
modify the portfolio's duration by an amount that will not necessarily
correspond to the steady and natural decline of the investor's horizon. On
the one hand, if interest rates do not change, the simple passage of a year,
for instance, will reduce duration of the portfolio by less than one year;
the money manager will have to change the composition of the portfolio
so that duration is reduced by a whole year. But, changes in interest rates
will also modify the portfolio's duration; as we saw in chapter 3, a de-
crease in interest rates increases duration, and higher interest rates entail
lower duration. Hence, a decrease of interest rates from one year to an-
other will reinforce the necessity of readjusting the portfolio so that the
portfolio's duration matches the investor's horizon.
There are a few caveats: the immunization procedure works well only
if all interest rates shift together (in other words, only if the interest rate
structure undergoes a parallel shift). And, as usual, investing in a bond
portfolio and rebalancing it for duration adjustment to the investor's needs
requires that the market for those bonds be large and highly liquid.
At the end of chapter 2, we encountered a remarkable property: in the
case of either a drop or a rise in interest rates, when the horizon was prop-
erly chosen, the horizon rate of return for the bond's owner was about the
same as if interest rates had not moved. In chapter 2, we hinted that this
horizon was the duration of the bond; this is why we have devoted a
chapter to dening this concept and describing its fundamental properties.
We will now demonstrate rigorously why duration does indeed, in some
circumstances that still need to be dened, protect the investor against
moves in rates of interest. We will call `ìmmunization'' the process by
which an investor can protect himself against interest rate changes by
suitably choosing a bond or a portfolio of bonds such that its duration is
equal to his horizon.
We will proceed as follows. In the rst section, we will introduce the
``highly probable'' existence of such an immunizing horizon, thus rein-
144 Chapter 6
forcing the rst insight we had earlier. In the second section, we will
demonstrate rigorously the immunization theorem.
6.1 An intuitive approach
Suppose that today we buy a 10-year maturity bond at par and that the
structure of spot rates is at and equal to 9%. Then tomorrow the interest
rate structure drops to 7%, and we do not see any reason why it would
move from this level in the coming years. Is this good or bad news? From
what we know, we can say that everything depends on the horizon we
have as investors. If our horizon is short, it is good news, because we
will be able to exploit the huge capital increase generated by the drop
in interest rates. At the same time, we will not worry about coupon rein-
vestment at a low rate, because this part of the return will be negligible in
the total return. On the other hand, if we have a long horizon, we will not
benet from a major price appreciation (because the bond will be well on
its way back to par), and we will suer from the reinvestment of coupons
for a long time at lower rates. So our conclusion is that with a short ho-
rizon it is good news, but with a long horizon it is bad news, and for some
intermediate horizon (duration of course, what else?) it may be no news,
because the change in interest rates will have very little impact, if any, on
the investor's performance (the capital gain will have nearly exactly
compensated for the loss in coupon reinvestment).
It is time now to recall our formula of the horizon rate of return. We
will immediately see why the problem we are addressing is not trivial.
From chapter 3, equation (5), we have
r
H
=
B(i)
B
0
_ _
1aH
(1 i ) 1
This expression depends on two variables, i and H. It is essentially the
product of a decreasing function of i (B(i)) and an increasing one (1 i).
Note that the decreasing function of i is taken to the power 1aH. So the
resulting function r
H
is a rather complicated function of i, with H playing
the role of a parameter inversely weighting the decreasing function of i.
In order to have a clear picture of the possible outcomes open to the
investor, we will rst consider all dierent possible scenarios for an in-
vestor with a short horizon, and all possible moves of the interest rate. We
145 Immunization: A First Approach
will then progressively lengthen the investor's horizon and see what
happens.
Consider rst a short horizon (one year or less). We already know that
a drop in interest rates is good news for our investor; symmetrically, it is
bad news if the rates increase: he will suer from a severe capital loss and
will not have the opportunity to invest his coupons at higher rates, since
he has such a short horizon. This one-year horizon rate of return is
depicted in gure 6.1.
Suppose now that our investor's horizon is a little longer (for instance
two years). How will his two-year horizon rate of return look if interest
rates drop? It will be lower than previously (in the one-year horizon case),
for two reasons:
i i
0
r
H
i
0
r
H =
= i
r
H =
= i
r
H
= 1
r
H
= 1
r
H
= 2
r
H
= 2
r
H
= 4
r
H
= 4
r
H
= 10
r
H
= 10
r
H
= D
Figure 6.1
The horizon rate of return is a decreasing function of i when the horizon is short and an
increasing one for long horizons. For a horizon equal to duration, the horizon rate of return rst
decreases, goes through a minimum for i Fi
0
, and then increases by i.
146 Chapter 6
1. The capital appreciation earned from the bond's price rise will be
smaller than before, since the price will be closer to par. Indeed, after one
year the bond will already have started going back to par; but after two
years, it will be farther down in this process. (From chapter 2, we know
that each year the bond drops by
hB
B
= i
1

c
B
[`0 in our case], where
i
1
is the new value of the rate of interest immediately after its change
from i
0
.)
2. The investor now has the worry (not to be taken too seriously, how-
ever) of reinvesting one coupon at a lower rate than in the one-year
horizon case.
We can be sure then that for these two reasons the two-year horizon
return will be below the one-year horizon return. Similarly, if interest
rates rise, the two-year horizon return will be above the one-year horizon
return for these two symmetrical reasons:
1. The capital loss will be less severe with the two-year horizon than with
the one-year, since the bond's value will have come back closer to par
after two years than after one year.
2. The two-year investor will be able to benet from coupon reinvestment
at a higher rate, which was not the case with the one-year investor.
Therefore, the curve depicting the two-year horizon rate will turn
counterclockwise around point (i
0
Y i
0
) compared to the one-year horizon
rate of return curve. More generally, we can state that all horizon return
curves will go through the xed point (i
0
Y i
0
), where i
0
represents the initial
rate of interest, before any change has taken place; in other words, what-
ever the horizon, the rate of return will always be i
0
if i does not move
away from this value. Why is this so? The answer is not quite obvious and
deserves some attention. When the bond was bought, the interest rate
level was at i
0
. Therefore, the price was B(i
0
) = B
0
. If the rate of interest
does not move, the future value of the bond, at any horizon H, is
B
0
(1 i
0
)
H
, and the rate of return r
H
transforming B
0
into B(1 i
0
)
H
is
such that B
0
(1 r
H
)
H
= B
0
(1 i
0
)
H
; it is therefore i
0
.1 Hence all curves
depicting the horizon rate of return as a function of i will go through
point (i
0
Y i
0
) in the (r
H
Y i) space, whatever the horizon, be it short or long.
1For an alternate demonstration of this property, see question 6.1 at the end of this chapter.
Now that we have seen that the curve depicting the slightly longer ho-
rizon rate of return would turn counterclockwise around point (i
0
Y i
0
), we
may wonder what the limit of this process would be if the horizon
becomes very large. If H tends toward innity, we can see immediately
that the horizon rate of return, r
H
=
B(i)
B
0
_ _
1aH
(1 i) 1 tends toward i.
So the straight line at 45
in (r
H
Y i) space is this limit. Consider now a long
horizon, perhaps 10 yearsthe maturity of the bond. If the interest rate is
lower than i
0
, the bond will be reimbursed at par, and all coupons will be
invested at a rate lower than i
0
. Obviously, the horizon rate of return will
be lower than i
0
. If the interest rate is higher than i
0
, coupons will be re-
invested at a higher rate than i
0
, and the horizon rate of return will be
higher than i
0
, increasing also by i.
We sense, however, that if the horizon is somewhat shorter (nine years,
for instance), then the curve will rotate clockwise this time, for similar
reasons to those we discussed earlier. If we consider the case i ` i
0
,
then the shorter horizon means the bond will still be above par, and the
coupons will suer reinvestment at a lower rate for a shorter time. For
these two reasons the return will be higher. If, on the contrary, i b i
0
, the
bond will still be somewhat below par, and the coupons will be reinvested
at the higher rate for a shorter time. Again, these two factors will entail a
lower rate of return.
We now have a complete picture of the behavior of the rate of return
for various horizons. We understand that the curve describing these
rates, as a function of the structure of interest rates, will denitely be
moving counterclockwise around the xed point (i
0
Y i
0
) when the horizon
increases, and we can well surmise that there might exist a horizon such
that the curve will be either at or close to at. That horizon does indeed
exist, and is none other than duration. Our purpose in section 6.2 will be
to demonstrate this rigorously.
6.2 A rst proof of the immunization theorem
We will now prove the following theorem: if a bond has duration D, and
if the term structure of the interest rates is horizontal, then the owner of
the bond is immunized against any parallel change in this structure if his
horizon H is equal to duration D. This theorem can be generalized to the
case of any term structure of interest rates, if the change in this structure is
a parallel one (we will prove this generalization in chapter 10).
148 Chapter 6
For the time being let us prove the simpler theorem above. What we
want to show is that there exists a horizon H such that the rate of return
for such a horizon goes through a minimum at point i
0
.
6.2.1 A rst-order condition
Let us recall that the rate of return at horizon H is such that
B
0
(1 r
H
)
H
= B(i)(1 i)
H
= F
H
(1)
where F
H
is the future value of the bond, or the bond portfolio, and
therefore r
H
is equal to
r
H
=
F
H
B
0
_ _
1aH
1 (2)
Minimizing r
H
is equivalent to minimizing any positive transformation
of it, and therefore is equivalent to minimizing F
H
. It suces then to
evaluate the derivative of F
H
with respect to i. We have
dF
H
di
=
d
di
[B(i)(1 i)
H
[ = B
/
(i)(1 i)
H
HB(i)(1 i)
H1
= 0 (3)
Multiplying (3) by (1 i)
1H
, we get
B
/
(i)(1 i) HB(i) = 0 (4)
We want
dF
H
di
= 0 to hold at point i = i
0
. Therefore, we must have
B
/
(i
0
)(1 i
0
) HB(i
0
) = 0 (5)
and
H =
(1 i
0
)
B(i
0
)
B
/
(i
0
) (6)
We have just demonstrated that the required horizon must be equal to
1i
0
B(i
0
)
B
/
(i
0
); one of the central properties of duration D, evaluated at i
0
, is
that it is precisely equal to this magnitude (see equation (3) in chapter 4).
Therefore, the horizon must be equal to duration at the initial rate of in-
terest (or return) for F
H
(and r
H
) to run through a minimum. The size of
this minimum is r
H=D
= i
0
, because the minimum occurs at i = i
0
, and for
that value of i, r
H
= i
0
.
6.2.2 A second-order condition
We have now a rst-order condition for r
H
to go through a minimum. In
order to ensure a second-order condition for a global minimum at i = i
0
,
it is relatively easy to prove that ln F
H
(i) has a positive second derivative.
This is what we will show now.
We have
ln F
H
= ln B(i) H ln(1 i) (7)
d ln F
H
di
=
d ln B(i)
di

H
1 i
=
1
B(i)
dB(i)
di

H
1 i
=
D
(1 i)

H
1 i
=
1
1 i
(D H) (8)
Equating this expression to zero leads to D = H (as before). The second
derivative of ln F
H
is
d
2
ln F
H
di
2
=
1
(1 i)
2

dD
di
(1 i) D H
_ _
Since D = H, we have
d
2
ln F
H
di
2
=
1
1 i
dD
di
(9)
We had shown previously that
dD
di
= (1 i)
1
S, where S is the variance
of the bond.
We nally get
d
2
ln F
H
di
2
=
1
1 i
[(1 i)
1
S[ =
S
(1 i)
2
and this expression is always positive, whatever the initial value i
0
. Con-
sequently F
H
and r
H
go through a global minimum at point i = i
0
when-
ever H = D, and the immunization proof in the simplest case is now
complete.
150 Chapter 6
Main formulas
.
Future value of bond (at term structure):
F
H
= B(i)(1 i)
H
.
Rate of return at horizon H:
r
H
=
F
H
B
0
_ _
1aH
1 =
B(i)
B
0
_ _
1aH
_ _
(1 i) 1
.
First-order condition for minimizing r
H
:
dr
H
di
= 0 =H =
1 i
0
B(i
0
)
B
/
(i
0
) = D(i
0
)
.
Second-order condition for minimizing r
H
:
d
2
r
H
di
2
b 0 =
S
(1 i)
2
b 0
Questions
6.1. Show that all curves r
H
= r
H
(i) for various horizons H (H =
1Y 2Y F F F Y y) go through point (i
0
Y i
0
) using an alternate demonstration.
In other words, show that (i
0
Y i
0
) is a xed point for all curves r
H
(i).
(Hint: Use formula (5) from section 3.4 of chapter 3.)
6.2. In our intuitive approach to section 6.1, would anything have
changed if we had assumed that initially we had bought the bond below
par or above par? (Hint: The answer is yes; be careful in your reasoning
by turning to an analytic argument.)
6.3. At the end of section 6.2, what is meant by ``the variance of the
bond''?
6.4. What are the measurement units of equation (9) in section 6.2.2?
7
CONVEXITY: DEFINITION, MAIN
PROPERTIES, AND USES
Earl of Kent. Seeking to give losses their remedies . . .
King Lear
Friar Francis. Let this be so, and doubt not but success
Will fashion the event in better shape
Than I can lay it down in likelihood.
Much Ado About Nothing
Chapter outline
7.1 Denition and calculation of convexity 155
7.2 Improving the measurement of a bond's interest rate risk through
convexity 157
7.3 Convexity of bond portfolios 160
7.4 What makes a bond (or a portfolio) convex? 162
7.5 Some important applications of duration, convexity, and
immunization 163
7.5.1 The future assets and liabilities problem: the Redington
conditions 164
7.5.2 Immunizing a position with futures 168
Overview
As we saw in chapter 1, the relationship between the interest rate and the
bond's price is a convex curve. In our usage of the term we refer not to the
convexity of this curve, but to the convexity of the bond itself. The more
convex a bond, the more rewarding it is to its owner, because its value will
increase more if interest rates drop and will decrease less if interest rates
rise.
The convexity of a bond depends positively on its duration and on its
dispersion. The dispersion of a bond is simply the variance of its times of
payment, the weights being the cash ows of the bond in present value.
It should be emphasized that substantial gains based on convexity can be
made on bond portfolios only in markets that are highly liquid and where
participants have not already attributed a premium to this convexity.
We have noted already, in chapter 2 that the value of a bond is a con-
vex function of the rate of interest, and we just saw in chapter 6 that du-
ration is very closely linked to the relative rate of change in the price of
the bond per unit change in interest rate (it was equal to
(1i)
B
dB
di
).
Consider now two bonds that have the same duration, for instance 6.99
years. This is the case for bond A, already encountered: it is valued at par,
oers a 9% coupon rate, and has a maturity of 10 years. If we want to nd
another bond, called B, that has the same duration and the same yield to
maturity (9%) as A, we can tinker with the other parameters that make up
the value of duration, namely the coupon rate and the maturity. For in-
stance, a lower coupon rate will increase duration, and it probably suces
to decrease maturity at the same time to keep duration at 6.99 years. This
is precisely what we will do: take a coupon rate of 3.1% and a maturity of
eight years for bond B; it will have a duration of 6.99. If the yield to
maturity is to be 9% for both bonds, the price of B will be way below
par, since its coupon rate is below i = 9%; in eect, its price will be 673
(rounding to the closest unit). The characteristics of bonds A and B are
summarized in table 7.1.
Consider now two investments of equal value in each bond: the rst
(portfolio :) consists of 673 bonds A bought for the amount of $673,000,
and the second (portfolio ) is made up of 1000 bonds B costing $673
eachan equivalent investment. At rst glance an investor could be in-
dierent as to which portfolio he chooses: they are worth exactly the same
amount, they oer the same yield to maturity, and they seem to bear the
same interest rate risk, since they have the same duration. However, the
investor would be wrong to believe this. If we calculate the value of each
portfolio if interest rates move up or down from 9%, portfolio : always
outperforms portfolio (see table 7.2). This is a surprising outcome, and
it is the direct result of the dierence in convexities between the two
bonds. In eect, the value of each portfolio can be pictured as in gure
7.1, investment in A (and bond A) being more convex than investment
in B (and bond B). The precise analysis of convexity is the purpose of
this chapter. In particular, we will ask what makes any bond more
convex than another one (as in this case) if they have the same duration,
Table 7.1
Characteristics of bonds A and B
Coupon
Rate Maturity Price
Yield to
Maturity Duration
Bond A 9% 10 years $1000
9% 6.99 years
Bond B 3.1% 8 years $673
154 Chapter 7
and how we can increase the convexity of a portfolio of bonds, given its
duration.
7.1 Denition and calculation of convexity
Convexity of a curve is always dened as the rate of change of its slope,
that is, by its second derivative. The more quickly the slope of a curve
changes in one direction (for instance, the steeper it becomes), the more
convex is the curve. For bonds it is customary to dene convexity as
follows:
Convexity =
1
B(i)
d
di
dB
di
_ _
I
1
B
d
2
B
di
2
Thus convexity is equal to the second derivative of B(i) divided by the
value B(i). The reason we divide the second derivative by B will soon
become clear.
Remember from chapter 4 on duration that we calculated the rst de-
rivative
dB
di
as
dB
di
=
T
t=1
(t)c
t
(1 i)
t1
(1)
Table 7.2
Value of portfolios and for various rates of interest
Portfolio : Portfolio
i (in %) Value of Investment in A (in $1000) Value of Investment in B (in $1000)
5 881 877
6 822 820
7 768 767
8 719 718
9 673 673
10 632 631
11 594 593
12 559 558
13 527 525
155 Convexity: Denition, Main Properties, and Uses
The second derivative of B with respect to i, divided by B, will then be
equal to
C =
1
B
d
2
B
di
2
=
1
B(1 i)
2
T
t=1
t(t 1)c
t
(1 i)
t
(2)
Before giving an example for the calculation of convexity, we should
notice immediately two important properties of convexity:
.
Property 1: A bond's convexity is always positive.
.
Property 2: The dimension of a bond's convexity is (years
2
). It is very
useful to remember this property, though it is often overlooked. It can be
reached either by the fact that the dimension of
1
B
d
2
B
di
2
is 1a(1at)
2
= t
2
or
by the fact that convexity is equal to the weighted average of all products
t(t 1), each of those having dimension (years
2
). The average is then
B A
A
B
Value of bond
i i
1
i
0
i
2
Figure 7.1
The eect of a greater convexity for bond A than for bond B enhances an investment in A
compared to an investment in B in the event of change in interest rates. Investment in A will
gain more value than investment in B if interest rates drop and it will lose less value if interest
rates rise.
156 Chapter 7
divided by the pure number (1 i)
2
; hence dimension of convexity is in-
deed (years
2
).
As an example, let us calculate the convexity of our bond A. Table 7.3
summarizes those calculations; the result is 56.5 (years)
2
.
7.2 Improving the measurement of a bond's interest rate risk through convexity
In chapter 4 on duration we saw how the change in the bond's value could
be approximated linearly, thanks to the duration. We will now see how
we can improve upon this with a second-order approximation. Writing
down the rst two terms of Taylor's series, we have
hB
B
e
1
B
dB
di
di
1
2
d
2
B
di
2
(di)
2
_ _
(3)
We will also consider here an increase of 1% in the rate of interest.
Hence we get di = 1% = 0X01. Earlier we had reached the part
1
B
dB
di
di
of this expression; this was equal to
1
1i
Ddi =
1
1X09
(6.995 years) (1%
Table 7.3
Calculation of the convexity of bond A (coupon: 9%; yield to maturity: 10 years)
(1)
Time of
payment
t
(2)
t(t 1)
(3)
Cash ow
in nominal
value c
t
(4)
Share of the
discounted cash
ows in bond's
value
c
t
(1 i)
t
aB
(5)
t(t 1) times share
of discounted cash
ows = (2) (4)
t(t 1)c
t
(1 i)
t
aB
1 2 9 0.0826 0.165
2 6 9 0.0758 0.456
3 12 9 0.0695 0.834
4 20 9 0.0638 1.275
5 30 9 0.0585 1.755
6 42 9 0.0537 2.254
7 56 9 0.0492 2.757
8 72 9 0.0452 3.252
9 90 9 0.0414 3.729
10 110 109 0.4604 50.647
Convexity: total of (5)
1
(1X09)
2
= 56X5 (years
2
)
per year) = 6X4% (a pure number). This approximation made use of
only the rst term of Taylor's expansion. If we wish to get a better ap-
proximation, we simply add the second term, equal to
1
2
1
B
d
2
B
di
2
di
2
=
1
2
[56X5(years)
2
[(1%)
2
a(year)
2
= 0X28%
(This is a pure number, of course. The reader can now see why convexity
was dened as (1aB) d
2
Badi
2
and also why convexity must be expressed
in (years)
2
. Otherwise it would not be compatible with Taylor's expansion
of
hB
B
. Since
hB
B
is a pure number, (di)
2
being expressed in 1/(years)
2
, con-
vexity
1
B
d
2
B
di
2
must have dimension (years)
2
.)
Consequently, the relative increase in the bond's value is given, in
quadratic approximation, by
hB
B
e
1
B
dB
di
di
1
2
1
B
d
2
B
di
2
(di)
2
= 6X4176 0X2825 = 6X135%
If, on the other hand, i decreases by 1%, we have, with di equal this time
to 1%,
= 6X4176 0X2825 = 6X700%
How good is the approximation now that we have gone to a second-order
one? In exact value, if the interest rate increases or decreases by 1%, the
bond's value will either go down by 6.14% or up by 6.71%, respectively.
The approximation is in fact highly improved by the use of convexity.
Table 7.4 summarizes these results.
In gure 7.2 we showed how well convexity improves the linear approx-
imation to a bond's value. We now give the relevant equations we used
Table 7.4
Improvement in the measurement of a bond's price change by using convexity
Change in the bond's price
Change in
the rate of
interest
in linear
approximation
(using duration)
in quadratic
approximation
(using both duration
and convexity) in exact value
1% 6.4% 6.135% 6.14%
1% 6.4% 6.700% 6.71%
158 Chapter 7
to draw this diagram, that is, the formulas of the straight line and the
parabola tangent to the curve of the bond's value, which are the linear
and quadratic approximations to the curve, respectively.
The linear approximation is given by the following equation, in space
(hBaBY di),
hB
B
0
=
1
B
0
dB
di
di = D
m
di (4)
Replacing hB by B B
0
and di by i i
0
, we have in space (BY i), after
simplications,
B(i) = B
0
[(1 D
m
i
0
) D
m
i [ (5)
The quadratic approximation is given, in space (hBaBY di), by
hB
B
0
=
1
B
0
dB
di
di
1
2
1
B
0
d
2
B
di
2
(di)
2
= D
m
di
1
2
C(di)
2
(6)
or, equivalently, in space (BY i), by
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
0
2000
1500
1000
500
Rate of interest (in percent)
Bonds value
Quadratic
approximation
Linear
approximation
Figure 7.2
Linear and quadratic approximations of a bond's value.
B(i) = B
0
C
2
i
2
(D
m
Ci
0
)i 1 D
m
i
0

C
2
i
2
0
_ _
(7)
Note that, in space (hBaBY di), the minimum of the parabola has an
abscissa equal simply to modied duration divided by convexity (D
m
aC).
This means that if B(i) displays little convexity, to make our approxima-
tion, we essentially make use of a part of the parabola that is relatively far
from its minimum.
Most nancial services that provide the value of convexity for
bonds divide the value of
1
B
d
2
B
di
2
by 200 (in this case, convexity would be
1
B
d
2
B
di
2
1
200
= 0X28(years)
2
). The reason for this is that this gure can readily
be added to the term containing modied duration. Thus one can obtain
immediately the (nearly exact) gures of 6X12% and 6X68% just by
adding this ``modied convexity number'' to minus or plus the modied
duration 6.4. Sometimes authors also refer to convexity as the value
1
2
1
B
d
2
B
di
2
1
100
, for analogous reasons. In this text we will stick to the original
denition given by
1
B
d
2
B
di
2
.
7.3 Convexity of bond portfolios
Remember how convexity of a bond was dened: it was the second de-
rivative of B(i) =

T
t=1
c
t
(1 i)
t
, divided by B(i). In fact, we see from
this very denition that the cash ow c
t
received at time t can be made of
any sum of cash ows received at that time. Hence, the denition of con-
vexity of a bond will immediately extend to bond portfolios. For that
matter, we can also observe that any coupon-bearing bond can be con-
sidered a portfolio of bonds; for instance, it can be viewed as a series of T
zero-coupon bonds (corresponding to each of the T cash ows to be
received before and at reimbursement time). It can also be considered a
combination of zero-coupon bonds and of coupon-bearing bonds. There-
fore, it is not surprising that the convexity of a portfolio of bonds is
dened simply as the second derivative of the portfolio's value P, divided
by P:
Convexity of P(i) IC
P
I
1
P
i
d
2
P(i)
di
2
(8)
For simplicity, let us rst restrict the number of bonds to 2; we then
have
160 Chapter 7
P = N
1
B
1
N
2
B
2
P(i) = N
1
B
1
(i) N
2
B
2
(i) (9)
Taking the rst derivative of (9) with respect to i, we get
dP(i)
di
= N
1
dB
1
di
(i) N
2
dB
2
di
(i) (10)
The second derivative of P(i), divided by P(i), is
1
P(i)
d
2
P(i)
di
2
=
N
1
P(i)
d
2
B
1
di
2
(i)
N
2
P(i)
d
2
B
2
di
2
(i) (11)
Let us then multiply and divide each term in the right-hand side of (11) by
B
1
and B
2
, respectively:
1
P(i)
d
2
P(i)
di
2
=
N
1
B
1
P
1
B
1
d
2
B
1
di
2

N
2
B
2
P
1
B
2
d
2
B
2
di
2
IX
1
C
1
X
2
C
2
(12)
where C
1
and C
2
denote the convexities of bonds 1 and 2, respectively.
Therefore, the convexity of the bond portfolio is simply the weighted
average of the bonds' convexities, the weights being the shares of each
bond in the portfolio. In familiar terms, we could say that the con-
vexity of the portfolio is the portfolio of the convexities. We can therefore
write
C
P
= X
1
C
1
X
2
C
2
(13)
As the reader may well surmise, this extends to a portfolio of n bonds,
and the demonstration of this property is immediate. We have
P(i) =
n
j=1
N
j
B
j
(i) (14)
Taking the second derivative of (6) and dividing by P(i), we have
1
P(i)
d
2
P(i)
di
2
=
n
j=1
N
j
B
j
P(i)
1
B
j
d
2
B
j
di
2
=
n
j=1
X
j
1
B
j
d
2
B
j
di
2
(15)
and therefore we have
C
P
=
n
j=1
X
j
C
j
(16)
This property will be very useful when our aim is to enhance convexity, in
the case of either defensive or active bond management.
We should add here for convexity the same caveat as the one we
pointed out earlier in chapter 4 on duration. First, it would be incorrect to
calculate the convexity of a portfolio by averaging the convexities of
bonds with dierent yields to maturity. Moreover, we should not forget
that we are dealing here with AAA bonds, whose default risk is practically
nil. This is not the case with a number of corporate securities, not to
mention bonds issued in some emerging countries, either by governments
or private rms. For those kinds of bonds, neither duration nor convexity
could ever provide a meaningful guide for investment.
7.4 What makes a bond (or a portfolio) convex?
We are now ready to determine what factors make a bond or a portfolio
of bonds more or less convex.
In order to make the expression of convexity
1
B
d
2
B
di
2
appear, let us take
the derivative of D expressed in the form
1i
B
dB
di
and equate the result to
(1 i)
1
S. We can write
D =
1 i
B(i)
B
/
(i)
dD
di
=
B (1 i)B
/
B
2
_ _
B
/
(i)
1 i
B
B
//
(17)
We have already shown, in chapter 4 on duration, that
dD
di
is equal to
(1 i)
1
S, where S is the dispersion, or variance, of the terms of the
bond. Thus we can equate the right-hand side of (17) to (1 i)
1
S.
Hence, after simplication, we have
1
B
1 D [ [B
/
(i)
1 i
B
B
//
= (1 i)
1
S (18)
Denoting B
/
aB = Da(1 i) and B
//
aB = C, and multiplying the last
relationship by 1 i, we nally get
162 Chapter 7
D(D 1) (1 i)
2
C = S
which can be conveniently written as
C =
1
(1 i)
2
[S D(D 1)[ (19)
This expression is quite important. It reveals that convexity depends on
both dispersion and duration; the relationship with duration turns out to
be a particularly strong one, since it is quadratic. Thus, a portfolio will be
highly convex when, for any given duration, the variance of its times of
payment is high.
7.5 Some important applications of duration, convexity, and immunization
Protecting an investment against uctuations in interest not only is im-
portant to the individual investor but also is highly relevant to insurance
companies, banks, pension funds, and other nancial institutions. It is in
fact those institutions that initiated research in this area. The rst contri-
bution came from Frederik Macaulay, whose article ``Some Theoretical
Problems Suggested by the Movements of Interest Rates, Bond Yields,
and Stock Prices in the U.S. Since 1856'' appeared in 1938. The rst
application to immunization, appearing in 1952, is attributed to F. M.
Redington.1
In order to show the importance of immunization and the central role
of duration played in this process, we will take up two issues of consider-
able breadth. First, how should a pension fund, or an insurance company,
set up its assets portfolio in such a way as to be practically certain it will
be able to meet its payment obligations in the future? This issue was rst
addressed by Redington (see reference above); we will call it the future
asset and liabilities problem. Second, we will show the close links between
duration and hedging with nancial futures.
1Frederik Macaulay, ``Some Theoretical Problems Suggested by the Movements of Interest
Rates, Bond Yields, and Stock Prices in the U.S. Since 1856'' (New York: National Bureau
of Economic Research, 1938); F. M. Redington, ``Review of the Principle of Life Oce
Valuations,'' Journal of the Institute of Actuaries, 18 (1952): 286340. Some of those early
texts have been reprinted in G. Hawawini, Bond Duration and Immunization (New York:
Garland, 1982).
7.5.1 The future asset and liabilities problem: the Redington conditions
Suppose that a company can make reasonably accurate predictions about
its cash outows in the next T years, denoted L
t
, t = 1Y F F F Y T, and about
its inows A
t
, t = 1Y F F F Y T. Suppose as before that the interest term
structure is at, equal to i; the present value of these future liabilities and
assets are, respectively, L
0
and A
0
:
L =
T
t=1
L
t
(1 i)
t
(20)
and
A =
T
t=1
A
t
(1 i)
t
(21)
Initially, Redington assumes that the initial net value of the future cash
ows, N = A L, is zero. The rst question is, how should one choose
the structure of the assets such that this net value does not change in the
event of a change in interest rates? From our knowledge of duration we
know that a rst-order condition for this to happen is that the sensitivities
of L and A have to i matchequivalently, that their durations are equal.
First, we want N = A L to be insensitive to i; in other words, we want
the condition
dN
di
=
d
di
[(A
t
L
t
)(1 i)
t
[ = 0 (22)
to be fullled.
Equation (22) leads to
dN
di
=
1
1 i
T
t=1
t(L
t
A
t
)(1 i)
t
=
1
1 i
(D
L
L D
A
A)
=
A
1 i
(D
L
D
A
) = 0 (23)
(since A = D), where D
L
=

T
t=1
tL
t
(1 i)
t
aL and D
A
=
T
t=1
tA
t
(1 i)
t
aA. The rst-order condition is indeed that D
L
D
A
, as
we had surmised. (This is called the rst Redington property.)
A second-order condition can also be deduced relatively easily. Since
we do not want the net value of assets to become negative, N(i) must be
164 Chapter 7
such that for all possible values within an interval of i
0
, N remains posi-
tive. This will happen if N(i) is a convex function of i within that interval
(note that this is a sucient condition, not a necessary one). We know
that convexity of a function is measured essentially by the second deriva-
tive of the function (for bonds, we divided this second derivative by the
initial value of the functions, for reasons that we have seen). Here we have
to deal with A(i) L(i), the dierence between two functions; hence, to
be convex its second derivative must be positive. Since the second deriv-
ative of a dierence between two functions is simply the dierence of the
second derivative of each function, we infer that the second derivative of
the assets with respect to i must be larger than the second derivative of the
liabilities (
d
2
A
di
2
b
d
2
L
di
2
). This is called the second Redington condition.
From this chapter (section 7.4, equation (19)), we know that once dura-
tion is given, convexity depends positively on the dispersion S of the cash
ows. Hence a necessary condition for the second Redington condition to
be met is that the dispersion of the inows be larger than the dispersion of
the outows. Given our knowledge of duration and convexities for bonds,
this condition is not dicult to meet, so that the net position of the rm
will always remain positive in the advent of changes in interest rates.
As a practical example where not even the rst of Redington's con-
ditions was satised, we can cite the well-known disaster of the savings
and loan associations in the United States in the early 1980s. These asso-
ciations used to have inows (deposits) with short maturities (and dura-
tions); on the other hand, their outows (their loans) had very long
durations, since they nanced mainly housing projects. So the rst
Redington condition was not met; this situation spelled disaster. Every-
thing would have been ne if interest rates had fallen; unfortunately, they
climbed steeply, causing the net worth of the savings and loan associa-
tions to fall drasticallya hard-learned lesson.
As a direct application of Redington's conditions, let us see how the
riskiness of a nancial deal can be modied simply by maneuvering
duration of assets or liabilities.2 Suppose that a nancial institution is
investing $1 million in a 20-year, 8.5% coupon bond. This investment is
supposed to be nanced with a nine-year loan carrying an 8% interest
rate. Assuming that the rm manages to borrow exactly $1 million at that
2We draw here and in the next section from S. Figlewski, Hedging with Financial Futures for
Institutional Investors (Cambridge, Mass.: Ballinger, 1986), pp. 105111. We have changed
the examples somewhat.
rate, both assets and liabilities have the same initial present value (V
A
and V
L
, respectively); however, they have dierent exposures to interest
changes. Here, since we assume a at term structure, the riskiness of
assets and liabilities is measured by making use of their Macaulay dura-
tion (D
A
and D
L
, respectively). We have
hV
A
V
A
e
D
A
1 i
A
di
A
= 9X165 di
A
(24)
and
hV
L
V
L
e
D
L
1 i
L
di
L
= 6X095 di
B
(25)
Indeed, the durations of the assets and liabilities can be calculated using
the closed-form formula when cash ows are paid m times a year, and
when the rst cash ow is to be paid in a fraction of a year equal to 0
(0 ` 0 ` 1):
D =
1
i
0
N
m
(i caB
T
) (1 iam)
(caB
T
)[(1 iam)
N
1[ i
(26)
where
0 Itime to wait for the next coupon to be paidY as a fraction of a year
(0 0 1)
mInumber of times a payment is made within one year
N Itotal number of coupons remaining to be paid
(In our case, 0 =
1
2
; m = 2; N is 40 for the 20-year bond and 18 for the
nine-year loan made to the nancial rm.)
Note that the above formula generalizes well formula (11) of chapter 4,
which corresponds to the case where coupons were paid annually (m = 1)
and the remaining time until the next coupon payment was one year
(0 = 1); also N was equal to T, the maturity of the bond (in years). Note
also how this formula enables you to know immediately that whatever the
frequency of the coupons to be paid, duration tends toward a limit when
N goes to innity (when the bond becomes a perpetual or a consol). In the
last term of this expression (the large fraction), the numerator is an ane
function of N, while the denominator is an exponential function of N.
Therefore, its limit is zero when N y; duration then tends toward
1
i
0 when the bond becomes a perpetual. Not only can we determine this
166 Chapter 7
limit, but this limit makes excellent economic sense, and the reasoning is
exactly the same as the one we proposed in chapter 4. The time you have
to wait to recoup 100% of your investment is
1
i
, to which you have to add
0, the time for the rst payment to be made to you. As an application, you
can verify that if a bond pays a continuous coupon at constant rate c(t)
such that after one year you receive the sum ca constant(you have
_
1
0
c(t) dt =
_
1
0
c dt = c), then the duration of a consol paying a continuous
coupon is
1
i
0 =
1
i
.
In our example, the durations are
D
A
= 9X944 years
D
L
= 6X583 years
and the modied durations are
D
mA
= 9X165 years
D
mL
= 6X095 years
The conclusion we draw from these gures is the following: although
the net initial position of the nancial rm is sound (the investment is
fully funded), its risk exposure presents a net duration of D
A
D
L
= 3X361
years and a net modied duration of D
mA
D
mL
= 3X070 years. This has
important consequences. Since we have
hV
A
V
A
e
dV
A
V
A
= D
mA
di
A
and
hV
L
V
L
e
dV
L
V
L
= D
mL
di
L
the rm's net position for this project (denoted V
p
), equal to V
p
= V
A
V
L
,
is such that
hV
p
V
=
hV
A
V
A
hV
L
V
L
e(D
mA
di
A
D
mL
di
L
)
Suppose for the time being that i
A
and i
L
receive the same increment,
di (di = di
A
di
L
). We would have
hV
p
V
p
e(D
mA
D
mL
) di = 3X070 di
This implies that for this project the rm has a positive net exposure
measured by a net modied duration of 3.070 years. It means that, in
linear approximation, if interest rates increase by one percent (100 basis
points), the net value of the project will diminish by 3.070%; in other
words, the rm has a project that is equivalent, for instance, to buying a
zero-coupon bond with a maturity equal to 3.070 years.
There is nothing wrong with that if the nancial manager estimates that
interest rates have a good chance of falling. But what if he is not so sure
of that happening? On the other hand, if he wants to hedge his project
against a rise in interest rates, he may choose to bring this net modied
duration to zero by restructuring either his assets, his liabilities, or both.
More generally, should his forecasts of di
A
and di
L
dier, he will want to
bring the durations of his assets and his liabilities (D
A
and D
L
) to levels
such that the following equation is veried:
D
mA
di
A
= D
mL
di
L
In doing so, the nancial manager may want to pay special attention to
the convexity of his assets and liabilities as well. This is an issue we will
address in later chapters. Before dealing with it, however, we should re-
mind ourselves of some important points. First, immunization is a short-
term series of measures destined to match sensitivities of assets and
liabilities. As time passes, these sensitivities continue to change, because
duration does not generally decrease in the same amount as the planning
horizon. Second, whenever interest rates change, duration also changes.
There may be cases where duration should be decreased because of the
shortening of the horizon, and at the same time interest rates have gone
up in such a way that they have eectively reduced duration in the right
proportion, but this is of course an exceptional case, not the general rule.
Finally, we should keep in mind that until now we have considered at
term structures and parallel displacements of them. More rened duration
measures and analysis are required if we do not face such at structures,
as we will see in our next chapter, and particularly in the last two chapters
of this book, where we consider interest rates as following stochastic
processes.
7.5.2 Immunizing a position with futures
When the nancial manager is faced with some liquidity problems in the
bonds market, he may well nd a way out by adding a futures position to
168 Chapter 7
his portfolio. What he will want to do is to take a position of n
f
futures
contracts on T-bonds, for instance, such that the total duration of his
portfolio is zero. His exposure to interest rate risk would then be reduced
to nil, subject to the qualications we mentioned in our last section.
Suppose that f is the value of a futures contract; the value of his posi-
tion in futures is then the number of contracts n
f
multiplied by f, that is,
n
f
f . If the sensitivity of his portfolio of futures assets and liabilities has to
be zero, the number of contracts he must buy (or sell), n
f
, must be such
that the duration of his portfolio must be zero; if D
f
is the duration of a
futures contract, this is tantamount to the equation
D
f
n
f
F D
A
V
A
D
L
V
L
= 0 (28)
which leads to
n
f
=
D
L
V
L
D
A
V
A
D
f
F
(29)
Some attention simply must be paid to the determination of the futures
contract's duration D
f
, which is calculated in the following way:
.
The futures price F is the futures settlement price multiplied by the
relevant deliverable bond's conversion price (established in the United
States by the Chicago Board of Trade, as explained earlier). This gives
the adjusted price.
.
The relevant yield to maturity of the futures is calculated by using
the adjusted price and the deliverable bond's cash ows as viewed from
delivery date.
From the above formula giving n
f
, we can see that the position in
futures will be positive (negative) if and only if D
L
V
L
is larger (smaller)
than D
A
V
A
.
Main formulas
.
Convexity:
C =
1
B
d
2
B
di
2
=
1
(1 i)
2
T
t=1
t(t 1)c
t
(1 i)
t
aB
.
Quadratic approximation of relative change in bond's value:
hB
B
eD
m
di
1
2
C (di)
2
.
Convexity of bond portfolios (for bonds with same return to maturity):
C
P
=
n
j=1
X
j
C
j
.
Relationship between convexity, duration, and bond's dispersion (or
variance of times of payment S):
C =
1
(1 i)
2
[D(D 1) S[
Questions
7.1. Assuming that coupon payments are made once per year, conrm
one of the results of table 7.1, for instance that the convexity of a ten-year
maturity, 6% coupon rate is 62.79 (years
2
) when the bond's rate of return
to maturity is 9%.
7.2. Looking at gure 7.2, you will notice that the quadratic approxima-
tion curve of the bond's value lies between the bond's value curve and the
tangent line to the left of the tangency point and outside (above) these
lines to the right of the tangency point. The fact that a quadratic (and
hence ``better'') approximation behaves like this is not intuitive: we would
tend to think that a ``better approximation'' would always lie between the
exact value curve and its linear approximation. How can you explain this
apparently nonintuitive result?
7.3. In section 7.3, what, in your opinion, is the crucial assumption we are
making when calculating the convexity of a 2-bond or n-bond portfolio?
7.4. In section 7.4, make sure you can deduce equation (18) from equa-
tion (17).
170 Chapter 7
8
THE IMPORTANCE OF CONVEXITY IN
BOND MANAGEMENT
2nd Lord. And how mightily sometimes we make us
comforts of our losses!
All's Well that Ends Well
Chapter outline
8.1 Comparing the benets of holding two coupon-bearing bonds with
the same duration 172
8.1.1 The case of defensive management (immunization) 173
8.1.2 The case of active management 174
8.2 Example of a zero-coupon bond compared to a coupon-bearing bond 174
8.2.1 The case of defensive management 176
8.2.2 The case of active management 176
8.3 Looking for convexity in building a bond portfolio 176
8.3.1 Setting up a portfolio with a given duration and higher
convexity than an individual bond 177
8.3.2 Results of enhanced convexity 178
Overview
If the bond's market is highly liquid, the search for convexity can sub-
stantially improve the investor's performance. If we consider the two
basic styles of management, active and defensive, we will see that con-
vexity will help the money manager.
Consider rst the active management style. If bonds have been bought
with a short-term horizon and with the prospect that interest rates will
fall, the rate of return will be higher with more convex bonds or port-
folios. Suppose now that the interest rates move in the opposite direction
from what was anticipated and rise; the loss will be smaller for highly
convex bonds than for those that exhibit less convexity.
On the other hand, consider a defensive style of management, such as
an immunization scheme. In such a case, any movement in interest rates
that occurs immediately after the purchase of the portfolio will translate
into higher returns if the portfolio is highly convex.
Of course, the same caveats that we pointed to previously when dis-
cussing immunization apply here: convexity will denitely enhance a
portfolio's performance only in those markets that are highly liquid.
Our purpose in this chapter is to stress the role of convexity in any
bond's performance, in both defensive and active management styles. We
will rst consider two examples. The rst one will make use of two
coupon-bearing bonds; in the second we will deal with a coupon-bearing
bond and a zero-coupon. We will then indicate how we can enhance the
convexity of a bond portfolio.
8.1 Comparing the benets of holding two coupon-bearing bonds
with the same duration
Let us consider two types of bonds (I and II), diering by their maturity
(10 years for type I, 20 years for type II). Each type has various coupons,
ranging from zero to 15, with a return to maturity equal to 9%. The cor-
responding duration and convexity for each bond are presented in table
8.1. We notice that increasing the coupon decreases both duration and
convexity. We know why duration decreases; with regard to convexity, we
cannot say a priori whether it will increase or not with the coupon, be-
Table 8.1
Duration and convexity for two types of bonds
Type I bond Type II bond
Maturity: 10 years Maturity: 20 years
Duration
(years)
Convexity
(years
2
)
Duration
(years)
Convexity
(years
2
)
Coupon (c) D =
1i
B
dB
di
CONV =
1
B
d
2
B
di
2
D =
1i
B
dB
di
CONV =
1
B
d
2
B
di
2
0 10 92.58 20 353.5
1 9.31 84.34 15.9 251.5
2 8.79 78.02 13.81 215.98
3 8.37 73.02 12.59 188.85
4 8.03 68.96 11.78 170.8
5 7.75 65.6 11.21 158
6 7.52 62.79 10.77 148.4
7 7.32 60.38 10.44 140.94
8 7.15 58.31 10.17 134.98
9 6.99 56.50 9.95 130.10
10 6.86 54.9 9.76 126.04
11 6.76 53.49 9.61 122.61
12 6.64 52.23 9.48 119.7
13 6.55 51.10 9.36 117.12
13.5 6.50 50.58 9.31 115.97
14 6.46 50.08 9.29 114.89
15 6.38 49.16 9.18 112.93
172 Chapter 8
cause the eect on dispersion is undetermined. Since convexity is a posi-
tively sloped function of both duration and dispersion, there is no way to
tell except to do the cumbersome exercise of the comparative statistics on
convexity. It turns out that the duration factor in convexity will generally
outperform the dispersion factor; therefore, convexity will diminish with
an increase of the coupon.
Let us now consider two bonds, one from each family, with the same
duration but diering in their convexity. Both bonds, A and B, have the
same duration (9.31 years). Bond A has a coupon of 1 and a convexity of
84.34 years
2
; bond B has a coupon of 13.5 and a convexity of 115.97
years
2
. Table 8.2 summarizes the characteristics of both bonds.
Bond A, of course, will be way below par (48.66), and bond B will be
above par (141.08). Their durations are the same, but their convexities are
not. This implies signicantly dierent rates of return, in both defensive
and active management.
8.1.1 The case of defensive management (immunization)
Suppose that we want to protect our investor against parallel changes in
the interest rate structure, this structure being at. Should the investor's
horizon be 9.3 years (at least for part of his portfolio!), we know that he is
protected by purchasing either A or B. But the fact that B exhibits more
convexity has a number of advantages that will become immediately
apparent. Suppose that initial rates are 9% and that they quickly move
up by 1 or 2%or that they drop by the same amount, staying there for
the remaining investment period. Table 8.3 indicates the 9.31 horizon rate
of return for both bonds.
The reader can see that the order of magnitude of the increase in the
rate of return over 9% for the more convex bond (B) is tenfold that of the
bond with lower convexity. Suppose, as a matter of illustration, that $1
million is invested in each bond. Should the rates of interest decrease to
7%, the future value of each investment would be:
Table 8.2
Characteristics of bonds A and B
Bond A Bond B
Coupon 1 13.5
Duration 9.31 years 9.31 years
Convexity 84.34 years
2
115.97 years
2
173 The Importance of Convexity in Bond Management
.
investment in A: 1,000,000(1 0X9008)
9X31
= 2,232,222
.
investment in B: 1,000,000(1 0X9085)
9X31
= 2,246,245
which implies a dierence of $14,023 for no trouble at all, except looking
up the value of convexity.
8.1.2 The case of active management
The vast majority of investment decisions in bonds are still made in an
active management framework. If interest rates are forecast to decrease,
managers will take a long position in bonds, hoping for a quick rise in the
bond's value. The drawback, of course, is an error in the forecast, imply-
ing a loss in value of the investment. What we want to illustrate here is
how convexity can enhance the performance of such a management style,
if successful, and protect the investor slightly by cushioning a bit the loss
incurred if a forecasting error has been made. Table 8.4 shows the 1-, 2-,
and 3-year horizon rate of return in the same scenarios considered earlier.
The results are quite spectacular, and even more so when the horizon is
shorter. For instance, an increase of nearly one percentage point in the one-
year performance rate can be obtained if bond B is chosen rather than bond
A. And in the unhappy circumstance of an increase in interest rates from
9% to 11%, the convexity of B will cushion the loss from 6.2% to 5.7%.
8.2 Example of a zero-coupon bond compared to a coupon-bearing bond
Let us now compare the following two investments. Both bonds have the
same duration. The dispersion of the zero-coupon, however, will be zero
(and its convexity will be limited to (1 i)
2
D(D 1)); on the other
hand, the dispersion of the coupon-bearing bond will be positive, and thus
its convexity will be higher, being equal to (1 i)
2
[D(D 1) S[. The
characteristics of our two bonds are summarized in table 8.5(a).
Table 8.3
Rates of return for A and B with horizon DF9.31 years when rates move quickly from 9% to
another value and stay there
Scenario
i = 7% i = 8% i = 9% i = 10% i = 11%
Bond A 9.008% 9.002% 9% 9.002% 9.008%
Bond B 9.085% 9.021% 9% 9.020% 9.079%
174 Chapter 8
Table 8.4
Rates of return when i takes a new value immediately after the purchase of bond A or bond B
(in percentage per year)
Scenario
Horizon (in years)
and rates of return
for A and B i = 7% i = 8% i = 9% i = 10% i = 11%
H = 1
R
A
27.2 17.1 9 1.0 6.2
R
B
28.1 17.9 9 1.2 5.7
H = 2
R
A
16.7 12.7 9 5.4 2.0
R
B
17.1 12.8 9 5.5 2.3
H = 3
R
A
8.9 8.9 9 9.1 9.1
R
B
8.9 8.9 9 9.1 9.2
Table 8.5
(a) Characteristics of bonds A and B
Bond A
(zero-coupon) Bond B
Coupon 0 9
Maturity 10.58 years 25 years
Duration 10.58 years 10.58 years
Convexity 103.12 years
2
159.17 years
2
(b) Rates of return of A and B when i moves from i F9% to another value after the bond has
been bought
Scenario
i = 7% i = 8% i = 9% i = 10% i = 11%
Bond A (zero-coupon) 9 9 9 9 9
Bond B 9.14 9.04 9 9.02 9.08
8.2.1 The case of defensive management
In the case of immunization against changes in interest rates, we would
get the results you see in table 8.5(b). Note, of course, that the horizon
rate of return of the zero-coupon B is perfectly horizontal and exactly
equal to 9% in every scenario, as opposed to the rate for the coupon-
bearing bond. This is due to the fact that if the zero-coupon bond is held
to maturity, its terminal value, B
T
, is completely independent of any
reinvestment of the coupon, since none occurred. The horizon T = D rate
of return, r
T
, is dened by B
0
(1 r
T
)
T
= B
T
; r
T
is then independent from
i and equal to r
T
= i
0
(9% in our example).
8.2.2 The case of active management
We sometimes hear that if interest rates are almost denitely going down,
the investor should have the longest maturity possible (always, of course,
in the range of defaultless bonds). In fact, the investor's portfolio's dura-
tion and convexity should be as high as possible. This means that between
A and B, the investor must choose B; he will therefore dismiss the zero-
coupon A. Table 8.6 shows that the zero-coupon is beaten by 1.7 per-
centage points if interest rates go down to 7% on a one-year horizon.
8.3 Looking for convexity in building a bond portfolio
Suppose we are unable to nd a bond with the same duration and higher
convexity as the one we are considering buying. We can easily solve this
Table 8.6
Rates of return when i takes a new value immediately after the purchase of bond A or bond B
(in percentage per year)
Scenario
Horizon and rates of
return for A and B i = 7% i = 8% i = 9% i = 10% i = 11%
H = 1
R
A
30.2 19.1 9 0.01 8.4
R
B
31.9 19.5 9 0.02 7.7
H = 2
R
A
18.0 13.4 9 0.001 0.08
R
B
18.8 13.6 9 4.9 1.2
H = 3
R
A
8.9 8.9 9 9.1 9.1
R
B
8.9 8.9 9 9.1 9.2
176 Chapter 8
problem by building a portfolio that will have the same duration but
higher convexity.
Before we start, it will be useful to remind ourselves of the following
three results, developed in the preceding chapters:
.
The duration of a portfolio is the portfolio of durations.
.
The convexity of a portfolio is the portfolio of convexities.
.
The convexity C of a bond, or a bond portfolio, is linked to its duration
D and to its dispersion S by the following formula:
C =
1
(1 i)
2
[D(D 1) S[
8.3.1 Setting up a portfolio with a given duration and higher convexity than an
individual bond
Consider a bond 1 with such characteristics that its duration is exactly 5
years and its convexity 28.6 years
2
(see table 8.7). Bonds 2 and 3 are such
that the equally weighted average of their durations is 5, and the same
average of their convexity is above that of bond 1, which is 28.6 years
2
; it
is in fact 48.73 years
2
.
We will thus form a portfolio having the same value and the same
duration as bond 1. (Since we describe the principle of building such a
portfolio here, we will not bother considering fractional quantities of
bonds. In order to avoid that problem, we would simply consider an
initial value equal, for instance, to 100 times the value of bond 1.)
Let N
2
and N
3
be the number of bonds 2 and 3 we will choose. Let B
1
,
B
2
, and B
3
be the respective present values of each bond. N
2
and N
3
must
be such that:
Table 8.7
Summary of duration and convexity values for bonds and portfolios
Price
Duration
(years)
Convexity
(years
2
)
Bond 1 105.96 5 28.62
Bond 2 102.8 1 1.75
Bond 3 97.91 9 95.72
Portfolio 105.96 5 48.73
.
The value of the portfolio is equal to B
1
:
N
2
B
2
N
3
B
3
= B
1
(1)
.
The duration of the portfolio is that of 1; thus the weighted average of
the durations D
2
and D
3
must be equal to D
1
. We must have
N
2
B
2
B
1
D
2

N
3
B
3
B
1
D
3
= D
1
(2)
Since D
1
= 5, D
2
= 1, and D
3
= 9, we must have
N
2
B
2
B
1
9
N
3
B
3
B
1
= 5
Let X
2
I
N
2
B
2
B
1
; 1 X
2
= X
3
I
N
3
B
3
B
1
. From X
2
9(1 X
2
) = 5, we get
X
1
= X
2
= 0X5 (as we could have gured out). We can then determine the
number of bonds N
2
and N
3
to be bought, as
N
2
= 0X5
B
1
B
2
= 0X5
105X96
102X8
= 0X5153
and
N
3
= 0X5
B
1
B
3
= 0X5
105X96
97X91
= 0X5411
More generally, of course, solving (1) and (2) simultaneously would
have given us directly the number of bonds as
N
2
=
B
1
B
2
D
3
D
1
D
3
D
2
_ _
(3)
N
3
=
B
1
B
3
D
1
D
2
D
3
D
2
_ _
(4)
Table 8.7 summarizes the duration and convexity of our bonds and
those of the portfolio that we have built up.
8.3.2 Results of enhanced convexity
Thus, while the single bond 1 with a duration of 5 years had a convexity
of about 29 years
2
, the portfolio with the same duration has more con-
vexity. This results solely from the fact that we have been able to get more
dispersion in the times of payment. As the reader may surmise, this
178 Chapter 8
enhanced convexity will result in higher values for the portfolio than for
bond 1, for any interest rate dierent from 7%, as table 8.8 shows.
Let us now compare the performance of the bonds. Suppose we have
an immunization policy, and therefore our horizon is duration, that is, 5
years. As table 8.9 shows, the increase in convexity signicantly increases
the performance of the portfolio.
As we observed before, the order of magnitude of the relative gain
above 7% is about tenfold for the portfolio compared to the single bond.
Suppose now that we are interested in short-term performance, with an
active management when interest rates are forecast to decrease. Table
8.10 clearly indicates that the performance of the portfolio is signicantly
better if interest rates do decrease and that losses are somewhat contained
if interest rates increase.
We can conclude from these observations that a natural way to increase
convexity is to try to replace any given portfolio by another one that will
exhibit more convexity. Formally, we could look for the shares X
i
Y F F F Y X
n
of a portfolio of n bonds that would maximize the convexity of the port-
Table 8.8
Values of bond 1 and of portfolio P for various values of interest rate
i B
1
(i) B
P
(i)
4% 122.28 123.48
5% 116.50 116.99
6% 111.06 111.18
7% 105.96 105.96
8% 101.16 101.25
9% 96.64 97.00
10% 92.38 93.15
Table 8.9
5-year horizon rates of return for bond 1 and portfolio
i r
1
H=5
r
P
H=5
4% 7.023% 7.230%
5% 7.010% 7.100%
6% 7.003% 7.025%
7% 7% 7%
8% 7.003% 7.024%
9% 7.010% 7.092%
10% 7.023% 7.203%
folio, C
P
, equal to
C
P
=
n
i=1
X
i
C
i
(where the C
i
's are the individual bonds' convexities) under the constraints
n
i=1
X
i
= 1Y X
i
0 (i = 1Y F F F Y n)
and, if we are using defensive strategy, a duration constraint
n
i=1
X
i
D
i
= D
where D is our chosen duration.
It would be rather simple to show that the solution would be to choose
a portfolio of two bonds, one with maximal duration and the other with
minimal duration. However, a number of problems are involved in such a
procedure. On the one hand, liquidity problems might arise when we
concentrate large investments on a restricted number of assets. On the
other hand, it might be costly to update the portfolio frequently, since one
of the bonds would be systematically short-lived. Nevertheless, keeping
an eye on the convexity of any bond portfolio is clearly a good idea. We
should add here that inasmuch as many investors share this point of view,
the possibility of convexity enhancing will entail arbitrage opportunities
and a price to pay for convexity.
Table 8.10
One-year horizon rates of return for bond 1 and portfolio
i r
1
H=1
r
P
H=5
4% 20.019% 21.194%
5% 15.443% 15.935%
6% 11.108% 11.225%
7% 7% 7%
8% 3.105% 3.203%
9% 0.590% 0.214%
10% 4.098% 3.294%
180 Chapter 8
Key concepts
.
Convexity measures how fast the slope of the bond's price curve
changes when the interest rate is modied.
.
Convexity can be rewarding for the bond's owner: the greater the con-
vexity, the higher the return if interest rates drop; the smaller the loss if
interest rates rise.
.
Dispersion of a bond is the variance of the times of payment of the
bond, the weights being the discounted cash ows (duration being the
average of those times of payment).
.
Convexity depends strongly and positively on duration (it increases with
the square of duration) and on the dispersion of the bond.
Basic formulas
.
Convexity I
1
B
d
2
B
di
2
=
1
(1i)
2
[D(D 1) S[
.
Dispersion IS =

T
t=1
(t D)
2
c
t
(1 i)
t
aB
Questions
These exercises can be considered as projects.
8.1. Assuming that bonds pay coupons once a year, determine the
duration and convexity values contained in tables 8.1 and 8.3. (Hint:
Remember that you can approach the exact values of duration and con-
vexity at any desired precision level by applying the approximations
D =
1 i
B(i)
dB
di
e
1 i
B(i)
hB
hi
and
C =
1
B
2
(i)
d
2
B
di
2
e
1
B
2
(i)
h
hi
hB
hi
_ _
with hi suciently small to obtain the desired precision level for D and C.
The advantage of this method is that you can rely solely on closed-form
formulas.)
8.2. Assuming that bonds pay coupons once per year, verify the results
contained in tables 8.7 through 8.10.
8.3. Consider the following bonds:
Bond A Bond B
Coupon rate 10% 5%
Par value $1000 $1000
(The data are from exercise 4.8 of chapter 4.)
a. Determine the convexities of each bond.
b. Suppose you have a defensive strategy, and that you want to immunize
the investor. What is each bond's rate of return at horizon H = D if in-
terest rates keep jumping from 12% to either 10% or 14%?
c. Consider now an active style of management, whereby your horizon is
one year only.
In conclusion, which bond would you choose?
182 Chapter 8
9
THE YIELD CURVE AND THE TERM
STRUCTURE OF INTEREST RATES
Wolsey. Your envious courses . . . no doubt in time will nd their t
rewards.
Henry VIII
Chapter outline
9.1 Spot rates 185
9.2 The yield curve and the derivation of the spot rate curve 187
9.3 Derivation of the implicit forward rates from the spot rate structure 192
9.4 A rst approach: individuals are risk-neutral; the pure expectations
theory 195
9.5 A more rened approach: determination of expected spot rates when
agents are risk-averse 201
9.5.1 The behavior of investors 201
9.5.1.1 Investors with a short horizon 201
9.5.1.2 Investors with a longer horizon 202
9.5.2 The borrowers' behavior 203
Overview
Interest rates vary according to the maturities of the debt issues they
correspond to. It is a well-known fact that, in the long run, the term struc-
ture of interest rates often increases with maturity; in other words, short-
term paper usually carries smaller interest rates than long-term bonds. Our
purpose in this chapter is twofold: we rst explain the generally increasing
structure of spot rates; we then turn to the more dicult question of
whether knowledge of the term structure enables us to infer anything
about the future spot rates expected by the market.
We start by reminding the reader of the denition of a yield curve: a
yield curve depicts the yield to maturity of bonds with various maturities.
This curve should not be confused with the term structure of interest
rates, dened by the spot rate curve. The yield to maturity of a bond is the
discount rate which, when applied to each of the bond's cash ows, would
make their total present value equal to the cost of the bond. As such, it is
the horizon return on the bond for the given maturity if all cash ows of
the bond (all coupons) can be reinvested systematically at a rate equal to
the yield to maturityand we know that there is no reason for this to
be the case. Thus, the concept of yield to maturity carries the same
drawbacks as the concept of the internal rate of return of an investment
project.
We then move on to show how the spot rate (term) structure of interest
rates can be calculated from the prices of coupon-bearing bonds of dier-
ent maturities. We then explain why, generally, this structure is increasing
with maturity.
From this spot rate structure, we may infer an implicit forward struc-
ture, that is, the structure of interest rates that would apply to contracts
that start in the future and have dierent maturities. Indeed, it is possible
to show that any combination of two spot rates (for two dierent matu-
rities) always implies a certain forward rate for a span of time equal to the
dierence between these maturities. Both an investor and a borrower can
always lock in a forward rate by using spot rates for dierent maturities.
For instance, suppose a one-year spot rate and a two-year spot rate. An
investor who wants to borrow in one year for one year can always do two
mutually osetting things today, implying no net cash disbursement for
him, which will give him what he wants. If today he borrows a given
amount for two years and at the same time he lends that same amount for
one year, he will not disburse anything today. On the other hand, he will
receive some cash in one year and will disburse another amount in two
years. This is tantamount to borrowing in one year, with a one-year
contract. We can easily determine the forward rate implied by such an
operation; indeed, such a forward rate is determined solely by the very
existence of spot rates for various maturities.
From that point we will try to answer a much more dicult question:
what are the reasons behind a given structure of interest rates? For in-
stance, suppose that a given structure is increasing with maturity at one
point in time. What are the reasons for this? Does it imply that the market
believes future spot rates will be increasing? As the reader may well sur-
mise, the answer to that question is far from obvious. Using reasonable
assumptions, we will show that a downward sloping term structure always
implies that the market expects future spot rates to fall. On the other
hand, an increasing structure does not necessarily mean that the market
expects future spot rates to rise.
Among the central concepts in bond analysis and management is the
term structure of interest rates and its implications with regard to possible
184 Chapter 9
future interest rates. Admittedly, this subject is not especially easy, and,
unfortunately, its diculty is compounded by a confusion often made
between the yield curve and the term structure of interest rates (the spot
rate curve). This is the reason we have chosen to present in this chapter
these two concepts, which are quite distinct. Our rst task will be to dene
them with precision. We will then show how the spot rate curve can be
derived from the yield curve. This will enable us to derive from the spot
rate curve the implied forward rate curve, and we will then ask whether
implied forward rates have anything to tell us regarding expected future
interest rates.
9.1 Spot rates
The spot rate is the interest that prevails today for a loan made today with
a maturity of t years. It will be denoted i
0Y t
. The rst index (0 here) refers
to the time at which the loan starts; the second index (t) designates the
number of years during which the loan will be running. We should warn
the reader that it is not unusual to encounter other notations for interest
rates bearing three indexes rather than our two. Sometimes interest rates
are referred to as i
0Y 0Y t
; the rst index (0) refers to the time at which
the decision to make the loan is made; the second index (0) refers to the
start-up time of the loan; and the third index (t) may be either the date at
which the loan is due or the number of years during which the loan will be
running. In this text, since we will practically always deal with interest
rate contracts that are decided today that come into existence either im-
mediately or at some future date, for simplicity we have decided to sup-
press the rst of the above three indexes.
The spot rate is, in eect, the yield to maturity of a zero-coupon bond
with maturity t, because i
0Y t
veries the equation
B
t
= B
0
(1 i
0Y t
)
t
(1)
where B
0
is the value today of a zero-coupon bond, t is its maturity, and
B
t
is its par (or reimbursement) value. We could also say, equivalently,
that i
0Y t
is the horizon rate of return of a zero-coupon bond bought today
at price B
0
and having an expectedand certainvalue of B
t
at time t.
From (1), we can extract the horizon rate of return, as we have done in
earlier chapters, and obtain
185 The Yield Curve and the Term Structure of Interest Rates
i
0Y t
=
B
t
B
0
_ _
1at
1 (2)
So the concept of spot rate is entirely equivalent to the concept of horizon
rate of return.
Dividing both sides of (1) by (1 i
0Y t
)
t
, we can write that i
0Y t
is such
that
B
0
=
B
t
(1 i
0Y t
)
t
(3)
We deduce from (3) that the spot rate is the internal rate of return of the
investment, derived from buying a bond B
0
and receiving B
t
t years later.
Alternatively, we can see that the spot rate i
0Y t
is the rate that makes the
net present value of that project equal to zero:
NPV = B
0

B
t
(1 i
0Y t
)
t
= 0 (4)
We conclude that the spot rate is the internal rate of the investment proj-
ect derived from buying the zero-coupon bond, or equivalently the yield
to maturity of the zero-coupon bond. The following table summarizes
these properties of the spot rate.
Equivalent Denitions
Spot rate i
0Y t
for a loan
starting at time 0 and in
eect for t years
.
Horizon rate of return of an investment
transforming an investment B
0
into B
t
t years
later:
B
t
= B
0
(1 i
0Y t
)
t
or i
0Y t
=
_
Bt
B
0
_
1at
1
.
Yield to maturity t of a zero-coupon bond
B
0
=
Bt
(1i
0Y t
)
t
.
Internal rate of return of a project where B
0
is
invested today and B
t
is recouped in t years:
NPV = B
0

B
t
(1 i
0Y t
)
t
= 0
186 Chapter 9
The term structure of interest rates is the structure of spot rates i
0Y t
for all
maturities t. It is important not to confuse this with the yield curve, which
we will dene in section 9.2. How can we know the spot rate curve i
0Y t
?
The easiest way to determine this information would be to calculate the
horizon rates of return for a whole series of defaultless zero-coupon
bonds; this would give us our result immediately. Suppose that B
0Y t
desig-
nates the price today of a zero-coupon bond that will be reimbursed at
B
t
in t years. We would simply have to calculate i
0Y t
= (B
t
aB
0Y t
)
1at
1 for
the whole series of observed prices B
0Y t
, and we would be done. Unfortu-
nately, markets do not carry many zero-coupon bonds, especially in the
range of medium to long maturities. Therefore, we have to infer, or cal-
culate, the spot rates by relying on the values of coupon-bearing bonds.
This is what we will do in the following section.
9.2 The yield curve and the derivation of the spot rate curve
Let us rst give a precise denition of the yield curve: it is the curve rep-
resenting the relationship between the maturity of defaultless coupon-
bearing bonds and their yield to maturity. In other words, it is the
geometrical depiction of the various yields to maturity corresponding to
the maturities of coupon-bearing bonds. This yield curve is easy to deter-
mine and widely published, because it is necessary simply to consider a set
of coupon-bearing bonds with various maturities and to calculate their
yields to maturity. However, this is not the term structure of interest rates,
that is, the structure which is important to us and from which we want to
infer the rates with which we can discount future cash ows or from which
we can infer what the market thinks of future interest rates.
The following scheme summarizes the denitional and informational
content of the yield curve:
Price of a series of defaultless,
coupon-carrying bonds with
maturity t (t = 1Y F F F Y n)
Yield of bond with maturity

t (t = 1Y F F F Y n)
|
Yield curve for any maturity
t (t = 1Y F F F Y n)
It is clear that the yield curve is very dierent from the term structure of
interest rates, that is, the spot rate curve. The yield curve does not convey
more information than its contents, which is the yield to maturity, and we
know that the drawbacks of this concept are entirely similar to the inter-
nal rate of return of an investment. This return is a horizon return only if
all reinvestments are made at a constant rate equal itself to the internal
rate of returnwhich has reason not to be true.
So we must try to deduce the spot rate curve from the prices of the
bonds with dierent maturities, rather than simply limiting ourselves to
the yield curve. We now present a method that enables us to do this.
Consider rst the one-year spot rate, t
0Y 1
. We will be able to calculate it
immediately from the price of a zero-coupon bond maturing in one year
or from a coupon-bearing bond also maturing in one year. Suppose that
our bond bears a coupon of 10 and that the bond's price is $99. The spot
rate i
0Y 1
must then be such that
99(1 i
0Y 1
) = 110
and thus
i
0Y 1
=
110
99
1 = 0X111 = 11X1%
Let us now consider the price of a coupon-bearing bond with a matu-
rity equal to 2 years, B
0Y 2
. Its price is equal to the present value of all cash
ows it will bring to its holder, discounted by the spot rates. One of them,
i
0Y 1
, is already known; the second one, i
0Y 2
, is the unknown, which we can
determine from the following equation:
B
0Y 2
=
10
1 i
0Y 1
110
(1 i
0Y 2
)
2
(5)
where, of course, the one-year spot rate i
0Y 1
is known to be equal to
11.1%. Suppose, for instance, that c = 10 and B
0Y 2
= 97X5. Then i
0Y 2
is
determined by
97X5 =
10
1X11

110
(1 i
0Y 2
)
2
and so
i
0Y 2
= 11X5%
Suppose now that a third bond has a maturity of three years. Its cou-
pon may very well be dierent from 10, but in our example, we will keep a
188 Chapter 9
coupon of 10. The corresponding spot rate i
0Y 3
will be determined by the
following equation:
96 =
10
1X11

10
(1X115)
2

110
(1 i
0Y 3
)
3
which leads to
i
0Y 3
= 11X7%
This process can be repeated for all other spot rates, if we have the
necessary information about a defaultless bond's price with the relevant
maturity. If we want to calculate the spot rate i
0Y n
, knowing all spot
rates i
0Y 1
Y F F F Y i
0Y n1
and the bond's value B
0Y n
, then i
0Y n
must verify the
relationship
B
0Y n
=
c
1 i
0Y 1

c
(1 i
0Y n1
)
n1

c B
nY 0
(1 i
0Y n
)
n
(6)
where B
nY 0
denotes the value, at year n, of a bond whose maturity is 0
(that is, the par value of the bond). This relationship thus implies that
i
0Y n
=
c B
nY 0
B
0Y n
c
1
1i
0Y 1

1
(1i
0Y n1
)
n1
_ _
_
_
_
_
1an
1 (7)
There is, of course, a more elegant way of presenting this derivation of
the spot rate structure. If we use the concept of a discount rate as a price,
we simply have to solve a system of linear equations.
Suppose indeed that we have n bonds at our disposal, each with matu-
rity T = 1Y F F F Y n. Denote by B
(t)
the value of the bond with the tth
maturity and c
(t)
1
Y F F F Y c
(t)
t
the cash ows corresponding to bond t; c
(t)
t
is
always the par value of the tth bond plus the last coupon. Also, denote, as
we did in chapter 2, the discount factor 1a(1 i
0Y t
)
t
as the price b(0Y t) of
one dollar received at time t in terms of dollars of year zero. We then have
the system of n equations in n unknowns b(0Y 1) F F F b(0Y n):
B
(1)
= c
(1)
1
b(0Y 1)
B
(2)
= c
(2)
1
b(0Y 1) c
(2)
2
b(0Y 2)
B
(3)
= c
(3)
1
b(0Y 1) c
(3)
2
b(0Y 2) c
(3)
3
b(0Y 3)
F
F
F
F
F
F
F
F
F
F
F
F
B
(n)
= c
(n)
1
b(0Y 1) c
(n)
2
b(0Y 2) c
(n)
3
b(0Y 3) c
(n)
n
b(0Y n)
Using the following notation:
BIthe vector of the bonds' values
CIthe (diagonal) matrix of the bonds' cash flows
b Ithe (unknown) discount factors vector
we have
B = Cb
which generally can be solved in b as
b = C
1
B (8)
For instance, in our former simple case, matrix C and vector B would be
C =
110 0 0
10 110 0
10 10 110
_
_
_
_
Y B =
99
97X5
96
_
_
_
_
Vector b is then
b =
0X9
0X8045
0X7178
_
_
_
_
Remember now that i
0Y t
is obtained through
i
0Y t
=
1
b(0Y t)
_ _
1at
1
We then have the spot rate structure
i
0Y 1
= 11X1%
i
0Y 2
= 11X5%
i
0Y 3
= 11X7%
190 Chapter 9
Let us now consider a more realistic example, where we try to deter-
mine a 10-year term structure, that is, the set of all spot rates spanning
horizons from one to 10 years. Table 9.1 summarizes the information we
have about the maturity, the coupon, and the equilibrium prices of a
series of 10 bonds paying their coupons on an annual basis. Each of these
bonds is considered as having a $100 par value.
Table 9.1 provides all the ingredients needed to construct the diagonal
matrix C whose inverse has to be multiplied by vector B of the bonds'
prices to obtain the zero-coupon price vector b. From this the spot rates
can be deduced through i
0Y t
= [b(0Y t)[
1at
1. The results are in table 9.2.
The spot rate interest structure we get is exactly the one we presented as
an example in chapter 2 (section 2.4, table 2.5).
Table 9.1
Maturity, coupon, and equilibrium price for a set of bonds
Maturity t (years) Coupon Equilibrium price vector B
1 5 100.478
2 5.5 101.412
3 6.25 103.591
4 6 103.275
5 5.75 102.501
6 5.5 101.199
7 4 92.220
8 4.5 94.247
9 6.5 107.434
10 6 104.213
Table 9.2
Resulting discount factors (zero-coupon prices) and spot rate structure
Maturity t (years) Discount factor b(0Y t)
Spot rate interest term
structure i
0Y t
1 0.95693 0.045
2 0.91136 0.0475
3 0.86507 0.0495
4 0.81957 0.051
5 0.77609 0.052
6 0.73355 0.053
7 0.69202 0.054
8 0.65408 0.0545
9 0.61763 0.055
10 0.58543 0.055
Simple and appealing as this method may look, it is unfortunately dif-
cult to apply directly. Indeed, observations may be more than complete
in the sense that many coupon bonds mature on each coupon date. In the
past 30 years, researchers have tried to cope with this problem with more
and more rened econometric methods. One could even say that they
have displayed remarkable ingenuity in their pursuit. However, we will
have to wait until chapter 11, where we deal with continuous spot and
forward rates of return, to present the latest research in this eld.
9.3 Derivation of the implicit forward rates from the spot rate structure
It is now natural to ask the following question: is it possible to exploit our
knowledge of the term structure of interest rates (the spot rate curve) to
calculate the implicit forward rates, and then to try to deduce from those
the future spot rates that are expected by the market? For instance, sup-
pose the current structure is increasing, with the long rates being higher
than the short rates. Does this reveal any information about future spot
rates that are expected by the market? In this example, does it mean that
the market believes that the future short rates will be higher than the
current short rates? As we shall see, the answer to such a question is far
from obvious. We will proceed in two steps: rst, we will suppose that
individuals are risk-neutral; then we will suppress this rather stringent
hypothesis and suppose that the majority of agents are risk-averse.
We should rst recognize that if, on any given market, there exists a
term structure of interest rates, this implies the existence of a forward
structure. The following example will make this clear. Suppose we know
the two spot rates corresponding to the one-year and two-year contracts,
denoted as i
0Y 1
and i
0Y 2
. From this structure, any individual can construct
today a one-year forward loan starting in one year; in other words, if an
individual wished to borrow a certain amount in one year, he can lock in
today a rate of interest that will be prevailing with certainty in one year,
by doing the following:
.
He borrows $1 today for two years and repays (1 i
0Y 2
)
2
in two years.
.
He lends $1 today for one year; he will therefore receive 1 i
0Y 1
in one
year. Hence, since he receives 1 i
0Y 1
in one year and will have to reim-
burse (1 i
0Y 2
)
2
a year later, in eect he will have paid an (implicit) for-
ward rate, applicable in one year for a one-year time span. This is denoted
f
1Y 1
, such that it transforms 1 i
0Y 1
into (1 i
0Y 2
)
2
a year later.
192 Chapter 9
Thus, the implied forward rate f
1Y 1
must be such that
(1 i
0Y 1
)(1 f
1Y 1
) = (1 i
0Y 2
)
2
1 f
1Y 1
=
(1 i
0Y 2
)
2
1 i
0Y 1
(9)
which is equivalent to
f
1Y 1
=
(1 i
0Y 2
)
2
1 i
0Y 1
1 (10)
Of course, the process is perfectly symmetrical from the point of view of
a lender. Suppose someone wishes to lock in a forward rate for a loan he
wants to make in one year for one year. He can easily lend $1 today for
two years at rate i
0Y 2
, and he can borrow $1 for one year at i
0Y 1
. In one
year, he will repay his one-year loan 1 i
0Y 1
, and after two years he will
receive (1 i
0Y 2
)
2
. This implies that he will have made a loan in one year
such that the forward rate he will get will be equal to f
1Y 1
=
(1i
0Y 2
)
2
1i
0Y 1
1, as
before.
These conclusions can be generalized to any forward rate, given a
whole spot rate structure. For instance, consider the implicit forward in-
terest rate starting at t and valid for the n following years, f
tY n
. A bor-
rower can lock in this rate, for instance, by borrowing $1 today at rate
i
0Y tn
and lending the same $1 today for a period of t years at rate i
0Y t
. In t
years, our individual will pay back (1 i
0Y t
)
t
dollars, and n years later, he
will receive (1 i
0Y tn
)
tn
dollars. This amounts to transforming, at time
t, (1 i
0Y t
)
t
into (1 i
0Y tn
)
tn
n years later. Thus the implied forward
rate, f
tY n
, is such that
(1 i
0Y t
)
t
(1 f
tY n
)
n
= (1 i
0Y tn
)
tn
(11)
(1 f
tY n
)
n
=
(1 i
0Y tn
)
tn
(1 i
0Y t
)
t
or
f
tY n
=
(1 i
0Y tn
)
(tn)an
(1 i
0Y t
)
tan
1 (12)
Let us call 1 plus any interest rate the capitalization rate, which is
therefore the ratio of a capital one year hence over the capital invested
today, when a given rate of interest transforms the latter into the former.
(This capitalization rate is also called the dollar rate of return.)
The reader will notice from (11) that the long-term spot rate i
0Y tn
is
related to the shorter-term spot rate i
0Y t
, as well as the implicit forward
rate f
tY n
, by a time-weighted geometric average of their respective capi-
talization rates. From (11) we can write
1 i
0Y tn
= (1 i
0Y t
)
t
(tn)
(1 f
tY n
)
n
(tn)
(13)
which is indeed a geometric average of the short-term capitalization rate
1 i
0Y t
and the implicit forward capitalization rate, the respective weights
being the shares
t
tn
and
n
tn
in the total time span considered t n.
From one given spot rate structure, we have calculated all implicit for-
ward rate structures. First, we considered an increasing spot structure
(table 9.3; the same results were also shown in gure 9.1). We then sup-
posed that the term structure was ``humped,'' increasing in a rst phase
and then decreasing. These results are summarized in table 9.4 and gure
9.2.
The reader should notice that when we consider one spot structure, this
implies the existence of a series of implicit forward structures, which can
be classied in several ways. For instance, we have all forward rates that
start in one year, for loans with maturities ranging from one to nine years
(in our example). This is the second line of table 9.3. Next, the implicit
forward rates start in two years, for time spans between one and eight
years (the third line in our table). Finally, we have the implicit forward
rate that starts in nine years and ends one year later (the last line and last
gure of our table). Figure 9.1 represents the rst six lines of table 9.3,
and the same has been done in gure 9.2 for table 9.4. But we could also
visualize implicit forward rates under yet another classication. If we
consider all forward rates with a one-year maturity, starting at dierent
dates, we will see that these are all the rates located in the main diagonal
of our table. All forward rates starting at various years and corresponding
to maturities of t years would be those in the diagonal, which is located
t 1 spaces further to the right of the main diagonal. As an example, we
have shown in the table all forward rates with a maturity of four years,
which are in the diagonal located three spaces to the right of the main
diagonal.
194 Chapter 9
9.4 A rst approach: individuals are risk-neutral; the pure expectations theory
Let us consider two periods. Our problem is to try to know what the
market expects the future one-year spot rate to be in one year, that is, i
1Y 1
,
relying on our knowledge of the present spot rates, which we call the short
rate (i
0Y 1
) and the long rate (i
0Y 2
). We will suppose that
.
Lenders try to get the highest return for their loans.
.
Borrowers try to pay the smallest interest rate.
Two possibilities are oered to lenders as well as to borrowers. Either
lenders can invest on the two-year market or they can perform two suc-
Table 9.3
Implicit forward rates derived from a given time structure of interest rates
(increasing structure)
Time at
which
loan
starts Time at which loan ends (t n)
(t) 1 2 3 4 5 6 7 8 9 10
0 8.0 8.5 9.0 10.0 11.0 11.5 11.73 11.8 11.8 11.8 Spot rate structure
1 9.0 9.5 10.7 11.8 12.2 12.4 12.4 12.3 12.2
2 10.0 11.5 12.7 13.0 13.0 12.9 12.8 12.6
3 13.1 14.1 14.1 13.8 13.5 13.2 13.0
Implicit four-year
4 15.1 14.6 14.1 13.6 13.3 13.0
maturity forward
5 14.0 13.6 13.1 12.8 12.6 contracts
6 13.1 12.7 12.4 12.3
7 12.0 12.0 11.9
Implicit one-year
8 11.8 11.8
maturity forward
9 11.8 contracts
cessive operations: the rst on the one-year spot market and the second in
one year on the one-year spot market. In a symmetrical way, borrowers,
on a two-year period, can borrow on the two-year market, or alter-
natively, they can borrow twice for a one-year period. In the rst case
they will sell a two-year maturity bond; in the second case they will sell
two successive one-year bonds.
We will show that if lenders and borrowers are indierent with respect
to those two strategies, then the expected one-year rate for year 1, E[i
1Y 1
[,
must be linked to the spot rates i
0Y 1
and i
0Y 2
in the following manner:
(1 i
0Y 2
)
2
= (1 i
0Y 1
)(1 E[i
1Y 1
[) (14)
In other words, E[i
1Y 1
[ must be the implied forward rate f
1Y 1
we intro-
duced previously. Analogous to what we showed earlier, the two-year
capitalization rate 1 i
0Y 2
is the geometric average between 1 i
0Y 1
and
the expected capitalization rate 1 E[i
1Y 1
[; (13) implies indeed
1 i
0Y 2
= (1 i
0Y 1
)(1 E[i
1Y 1
[)
1a2
(15)
0 1 2 3 4 5 6 7 8 9 10 11
0.07
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
Time at which loan ends
S
p
o
t

r
a
t
e
s

a
n
d

i
m
p
l
i
c
i
t

f
o
r
w
a
r
d

r
a
t
e
s
Spot structure
1-year forward
5-year forward
4-year forward
3-year forward
2-year forward
Figure 9.1
Implicit forward rates derived from a given spot rates structure (increasing structure).
196 Chapter 9
Let us now show that (13) is valid under the risk-neutrality hypothesis
we have made. Suppose that (13) does not hold and that the expected rate
E[i
1Y 1
[ is such that it seems less protable for the investor to invest on the
two-year market than to lend twice consecutively on the one-year market.
This hypothesis implies that it will be less costly for the borrower to bor-
row on the two-year market than to try to raise funds twice on the shorter
term markets. Suppose our data are the following:
.
One-year spot rate = i
0Y 1
= 6%
.
Expected future spot rate, in one year = E[i
1Y 1
[ = 7%
The geometric average of (1 i
0Y 1
) and (1 E[i
1Y 1
[) is [(1X06)(1X07)[
1a2
= 1X064988 e1X065; the two-year spot rate that would make both lenders
and borrowers indierent between the two strategies would be i
0Y 2
=
1X065 1 = 6X5%. Note that for two fairly close gures, namely 1.06 and
1.07, the simple geometric mean is extremely close to the arithmetic
mean.) Let us then suppose that the two-year spot rate is signicantly
smaller than 6.5%, being equal to 6.2%. We would then have
0 1 2 3 4 5 6 7 8 9 10 11
0.05
0.11
0.1
0.09
0.08
0.07
0.06
Time at which loan ends
S
p
o
t

r
a
t
e
s

a
n
d

i
m
p
l
i
c
i
t

f
o
r
w
a
r
d

r
a
t
e
s
Spot structure
1-year forward
5-year forward
4-year forward
3-year forward
2-year forward
Figure 9.2
Implicit forward rates derived from a given spot rates structure (hump-shaped spot structure).
(1X062)
2
` (1X06)(1X07)
and therefore
(1 i
0Y 2
)
2
` (1 i
0Y 1
)(1 E[i
1Y 1
[)
Under these conditions, borrowers will want to enter the two-year mar-
ket, and they will leave the one-year market. Symmetrically, investors will
do just the opposite: they will certainly want to quit the two-year market
to invest in the one-year market. Here is a possible scenario, which we
have summarized in gure 9.3: the initial supply and demand of loanable
funds on the one-year and two-year markets are depicted in full lines. The
equilibrium rates are 6% and 6.2%, respectively, as we supposed earlier.
Table 9.4
Implicit forward rates derived from a given time structure of interest rates
(``humped'' shape)
Time at
which
loan
starts Time at which loan ends (t n)
(t) 1 2 3 4 5 6 7 8 9 10
0 7.0 8.2 8.8 9.0 8.5 8.0 7.75 7.6 7.5 7.5 Spot rate structure
1 9.4 9.7 9.7 8.9 8.2 7.9 7.7 7.6 7.6
2 10.0 9.8 8.7 7.9 7.6 7.4 7.3 7.3
3 9.6 8.0 7.2 7.0 6.9 6.9 6.9
4 6.5 6.0 6.1 6.2 6.3 6.5
Implicit four-year
5 5.5 5.9 6.1 6.2 6.5
maturity forward
6 6.3 6.4 6.5 6.8 contracts
7 6.6 6.6 6.9
Implicit one-year
8 6.7 7.1
maturity forward
9 7.5 contracts
198 Chapter 9
Since borrowers prefer to choose the two-year market rather than roll-
ing over a one-year loan, they will increase their demand for loanable
funds on the two-year market and will reduce their demand on the one-
year market. On the former market, their demand will move to the right,
and it will move to the left on the latter. As for investors, they will want to
stay away from the long rates and will start looking for the short rates.
Their supply of loanable funds will move to the left on the two-year
market and to the right on the one-year market.
We can readily see the consequences of these modications in the be-
havior of borrowers and lenders: the two-year spot rate will increase and
the one-year spot rate will diminish, under the conjugate eects of an ex-
cess demand for longer loans and an excess supply for shorter ones.
Note here that we do not know which new equilibrium will be ob-
tained. (We have only one equation (13) in two unknowns (i
0Y 1
and i
0Y 2
),
D
S
One-year market
6%
5.7%
i
0,1
D
S
D
S
Two-year market
6.35%
6.2%
i
0,2
D
S
Figure 9.3
If 6% and 6.2% are equilibrium rates, respectively, and if the expected spot rate E[i
1, 1
] is 7%,
we have the following inequality:
(1.062)
2
H(1.06) (1.07)
which entails arbitrage opportunities. It also becomes:
. less expensive for borrowers to go to the long market; they will enter the two-year market
and exit the one-year market;
. less rewarding for investors to be on market 2 than on market 1; they will want to exit
market 2 and enter market 1.
A new equilibrium can be reached with 5.7% on market 1 and 6.35% on market 2. We will
then have
1.0635
2
F(1.057) (1.07)
which does not allow arbitrage opportunities any more.
E[i
1,1
]
E[i
1,1
]
supposing that the market's forecast of the future spot rate E[i
1Y 1
[ is
not aected by the movements in the spot rates we have just described.)
The only thing we can be sure of in this model where investors and bor-
rowers are risk-neutral is that i
0Y 1
, i
0Y 2
, and E[i
1Y 1
[ must be related by (13).
For instance, i
0Y 2
could increase from 6.2% to 6.35%, and i
0Y 1
could drop
from 6% to 5.7%, as indicated in our graph (dotted lines). This would in-
deed lead to a new equilibrium, since
(1X0635)
2
e(1X057)(1X07)
Let us now remind ourselves of the property according to which the
two-year ratio is the geometric mean of the one-year capitalization ratio
and the one-year expected capitalization ratio. This implies that spot rates
and the expected future spot rate are those indicated in the diagram be-
low, if the two-year rate is higher than the one-year rate.
This implies that E[i
1Y 1
[ is higher than i
0Y 1
and thus that the market
believes that the short rate will increase. Consequently, when agents are
risk-neutral, an increasing term structure of interest rates indicates that
the market forecasts an increase in short rates.
Symmetrically, suppose that the term structure of interest rates is
decreasing. We then have the following situation:
Since E[i
1Y 1
[ is necessarily lower than i
0Y 2
, it is lower than i
0Y 1
. A decreas-
ing structure implies a forecast of falling short rates. Finally, if the term
structure is at (if i
0Y 1
= i
0Y 2
), this means that agents throughout the mar-
ket believe that short rates will stay constant.
We have shown how, in equilibrium and in a universe where agents are
risk-neutral, the expected one-year spot rate can be determined simply by
applying equation (13), or, equivalently, by writing
E[i
1Y 1
[ =
(1 i
0Y 2
)
2
1 i
0Y 1
1 (16)
which is exactly the implicit forward rate dened in (10).
200 Chapter 9
An obvious and serious problem arises, however, if we consider long
periods for lending or borrowing. We have supposed that all agents were
risk-neutral and that they accepted the arbitrage equations (12) and (15)
as if the forecast of future rates, i
1Y 1
and i
tY n
, respectively, did not convey
any sort of risk worth worrying about. In real life, this is certainly not the
case: those forecasts are highly error prone; they do carry some element of
risk; and the vast majority of agents are risk-averse. The time has now
come to abandon the risk-neutrality hypothesis and to evaluate the con-
sequences associated with risk-averse agents.
9.5 A more rened approach: determination of expected spot rates when agents are
risk-averse
We want to answer the following question: are lenders and borrowers
indierent between investing (borrowing) on the two-year market and
undertaking similar operations twice on the one-year market, the rst time
at a rate known with certainty (i
0Y 1
) and the second time at an expected
rate E[i
1Y 1
[? In other words, does an arbitrage relationship such as
(1 i
0Y 2
)
2
= (1 i
0Y 1
)(1 E[i
1Y 1
[) (17)
really hold, for investors as well as for borrowers? Earlier, we answered
yes to this question, because we supposed that whatever risk was involved
in using the estimate E[i
1Y 1
[, agents considered that bad surprises were
weighted exactly evenly against good surprises, which is what made the
agents neutral to risk. But this is an oversimplication of reality. We will
rst consider the case of investors. In this category, we will distinguish
those with a short horizon from those with a long horizon. We will
then turn to the case of the borrowers and examine the same type of
distinction.
9.5.1 The behavior of investors
9.5.1.1 Investors with a short horizon
The short route (consisting of making two one-year loans) is not risky;
however, the two-year route involves much more risk, because if investors
want to cash in their investment, they are exposed to a potential capital
lossalthough it is also possible that they will benet from a capital gain
if in the meantime interest rates have dropped.
To make matters simple, suppose that both spot rates i
0Y 1
and i
0Y 2
are
equal (for instance, both are equal to 9%). An investor can lend his money
for one year at 9%. Suppose that he lends at 9%, buying a bond that
might be at par, and that immediately afterward the interest rates move,
either up or down. If they move up, the value of this bond becomes the
dotted line below the horizontal (at height B
T
) in gure 9.4. After one
year, however, he can recoup his original capital (B
T
). If he then reinvests
his money, and if interest rates do not move in any signicant way, his
capital will not be aected. This is what we have described by the hori-
zontal dotted line between t = 1 and t = 2. Note, of course, the fact that
our investor will be able to roll over his loan at a higher rate, since we
have supposed that rates would not uctuate after their initial rise.
If rates go down, the short investor benets from a capital gain during
the rst year but none whatsoever in the second year, and his reinvest-
ment rate after one year will be supposedly lower than in the rst year.
9.5.1.2 Investors with a longer horizon
For those investors who choose the long two-year route, the possible
capital loss following a rise in interest rates is indicated by the continuous
line in gure 9.4, below B
T
. As the reader can immediately see, these
B
T
Potential gain
Potential loss
One two-year investment
Two one-year
investments
Time 2 1 0
Figure 9.4
Value of a single two-year investment and two one-year investments if the rate of interest is
modied immediately after the rst investment, at time 0. The two-year investment entails
additional prots and losses compared to the one-year investment. It is thus riskier. Those
potential losses, as well as the potential gains, are represented by the hatched area.
202 Chapter 9
potential losses last for the whole two-year time span, and their magnitude
is always higher than those for the investor who chose the short route.
Admittedly, the potential gains (indicated by the hatched area above the
horizontal line B
T
) are higher, but the risk is also considerably higher.
The advantages and the drawbacks of the short investments compared
to those of the long investment are summarized in table 9.5. From the
point of view of capital gains, the two-year investment is much riskier
than rolling over a one-year loan; but, of course, if we consider reinvest-
ment risk, the double loan is riskier, since our investor is liable to see his
coupon rate decrease if interest rates go down. It is thus clear that each
strategy carries a specic risk: the long investor fears a capital loss in case
of an increase in interest rates, while the short strategy involves the risk of
rolling over at a lower interest rate. It is obvious that the capital loss risk
is more substantial than the reinvestment risk. This will have important
consequences.
Indeed, we may conclude from this analysis that investors will be more
interested in looking for short deals rather than long ones; therefore, there
will be a natural tendency for supply to be relatively abundant on short
markets and relatively scarce on long ones. It remains for us now to see
whether the behavior of borrowers will reinforce these consequences or
not.
9.5.2 The borrowers' behavior
Borrowers with a short horizon will have a natural tendency to borrow on
short-term markets. They could of course borrow on the two-year market,
but they would have to redeem their bonds after one year, and we know
that capital gains or losses are at stake. Consequently, short-term bor-
rowers will try to minimize their risk by borrowing on the short market.
Borrowers with longer horizons will preferably choose the long mar-
kets. Of course, they could roll over a series of short loans, but they would
Table 9.5
Potential gains and losses of investors, corresponding to the long and short strategies
Capital gain or loss Reinvestment gain or loss
Strategy i increases i decreases i increases i decreases
Short investment
and roll-over
Small loss Small gain Reinvestment at
a higher rate
Reinvestment at
a lower rate
Long investment Large loss Large gain No reinvestment; same coupon
thus take the risk of an interest rate increase, while long markets do not
carry such risk. Thus, their behavior will tend to counter the consequences
of the short-term borrowers' action.
As before, we can ask the question about which tendency will prevail.
The answer depends on the magnitude of the loanable funds' demands on
each market. Generally speaking, we can say that borrowers have typi-
cally long horizons, and they will usually accept paying a risk premium.
Since their behavior will tend to increase the long-term rates compared to
the short-term ones, we can see that the consequences generated by the
lenders' behavior will be reinforced. In sum, on the long markets we will
generally nd a relatively scarce supply of funds and a relatively abundant
demand for funds. Symmetrically, short-term markets will usually witness
an ample supply and a short demand for funds, leading to long interest
rates that will often be higher than short ones.
The conclusion we can draw from this analysis is this: generally, both
suppliers and borrowers of funds will agree that a liquidity premium must
be paid on long loans. It will then be quite acceptable to admit that the
long-term capitalization rate is not a weighted average of the short-term
rate and the expected short-term rate but that it is somewhat higher than
that. Thus we have an inequality of the following type:
(1 i
0Y 2
)
2
b (1 i
0Y 1
)(1 E[i
1Y 1
[) (18)
We can therefore justify the existence of a ``risk premium'' to be received
by investors and paid by borrowers in the following way. It is simply
equal to the additional amount that has to be added to the expected spot
rate in order to make the square of the long capitalization rate equal to
the product of the two one-year capitalization rates. Thus, it will always
be such that
(1 i
0Y 2
)
2
= (1 i
0Y 1
)(1 E(i
1Y 1
) ) (19)
We can now draw some conclusions from the general shape of the in-
terest rates term structure. Suppose rst that this structure is at, that is,
i
0Y 2
= i
0Y 1
. Does this observation imply that the market forecasts that the
future spot rates will remain constant? We just showed that both lenders
and borrowers agreed that long investments should be rewarded by a risk
premium. Consequently, we can conclude from this that future spot rates
are expected by the market to go down. If i
0Y 2
= i
0Y 1
, equation (19) per-
mits us to write
204 Chapter 9
1 i
0Y 1
= 1 E[i
1Y 1
[ (20)
and thus
E[i
1Y 1
[ = i
0Y 1
(21)
The market therefore expects a decrease in the spot rate when the spot
rate structure is constant.
Let us suppose now that we face a decreasing structure (i
0Y 2
` i
0Y 1
). We
have, for instance, 1 i
0Y 2
= 0(1 i
0Y 1
), 0 ` 0 ` 1. Inequality (18) can
then be written, after simplifying by (1 i
0Y 1
),
0
2
(1 i
0Y 1
) b 1 E[i
1Y 1
[ (22)
and we have
1 E[i
1Y 1
[
1 i
0Y 1
` 0
2
` 1 (23)
Consequently, E[i
1Y 1
[ ` i
0Y 1
; the market expects the future spot rate to be
lower than today.
Let us examine nally what conclusions can be drawn if we face an
increasing structure of interest rates. Since long rates integrate a liquidity
premium, we are not in a situation to know, a priori, whether the struc-
ture that would not incorporate such a premium would be increasing or
not. As a result, we cannot deduce from such an increasing structure that
the market expects spot rates to rise. What we are able to say is that an
increasing structure corresponding to constant expected spot rates de-
nitely existsbut we do not know where exactly it is located, since we do
not know the exact magnitudes of all risk premiums. If the observed
structure is above this unknown error, then the market denitely expects
spot rates to rise. If it is under it, spot rates are expected to decrease.
We have summarized in table 9.6 the conclusions that can be drawn
from a set of observations of the term structure of interest rates when
we try to make our reasoning under two dierent types of hypotheses:
the pure expectations hypothesis, implying risk-neutrality of agents (a
hypothesis we are not very willing to accept); and the liquidity premium
hypothesis, corresponding to risk aversion of economic actors, a hypoth-
esis that is generally accepted. We need to stress, however, that even if we
can infer from an interest rate structure what the market thinks the future
interest rates will be, this is a far cry from being able to predict what will
Table 9.6
Relationship between observed term structures of interest rates and the implied forecasts of
future spot rates, according to the hypotheses of risk neutrality and risk aversion of economic
agents
i
Maturity
Interest rates are expected
to rise
Indeterminacy of expected
spot rates
to remain constant
to fall
to fall
to fall
Pure expectations theory
(risk-neutrality hypothesis)
Implied forecast of future spot rates Observed time structure
of interest rates
Liquidity premium theory
(risk-aversion hypothesis)
i
Maturity
i
Maturity
206 Chapter 9
actually happen in the future. Hence the importance of immunization, for
which we will demonstrate a general theorem in our next chapter.
Key concepts
.
Yield curve: the relationship between the maturity of defaultless
coupon-bearing bonds and their yield to maturity.
.
Term structure of interest rates: the spot rate curve, or the relationship
between the maturity of defaultless zero-coupon bonds and their yield to
maturity.
.
Forward interest rates: interest rates set up today for loans that will be
made at a future date for a well-determined time span.
.
Implied forward interest rates: the forward rates that are implied by the
current spot rate curve.
Main formulas
.
Spot rate (for rates compounded once per year):
i
0Y n
=
c B
nY 0
B
0Y n
c
1
1i
0Y 1

1
(1i
0Y n1
)
n1
_ _
_
_
_
_
1an
1 (7)
.
Discount factor:
b
0Y t
=
1
(1 i
0Y t
)
t
.
Vector of bonds values (B):
B = Cb
where
C = (diagonal) matrix of the bonds' cash flows
b = discount factors vector
.
Discount factors vector:
b = C
1
B (8)
.
Forward rate decided upon today for contract starting at t for a trading
period of n years:
f
tY n
=
(1 i
0Y tn
)
(tn)an
(1 i
0Y t
)
tan
1 (12)
Questions
9.1. Consider a spot rate i
0Y t
for a loan starting at time 0 and in eect for t
years. Can you give three other equivalent concepts of this spot rate in
terms of horizon rate of return, yield to maturity, and internal rate of an
investment project, respectively?
The next three questions can be considered as projects.
9.2. Using as input the data on the set of bonds given in table 9.1 (section
9.2), derive the discount factors and the spot rate interest term structure
i
0Y t
as contained in table 9.2 of the same section.
9.3. Determine the implicit forward rates derived from an increasing
structure of spot interest rates of your choosing.
9.4. Determine the implicit forward rates derived from a hump-shaped
structure of spot interest rates of your choosing.
208 Chapter 9
10
IMMUNIZING BOND PORTFOLIOS
AGAINST PARALLEL MOVES OF THE SPOT
RATE STRUCTURE
Earl of Salisbury. . . . much more general than
these lines import . . .
King John
Chapter outline
10.1 Value of a bond in continuous time for a given term structure 211
10.2 A fundamental property of duration 212
10.3 Value of a bond at horizon H 212
10.4 A rst-order condition for immunization 213
10.5 A second-order condition for immunization 214
10.6 The diculty of immunizing a bond portfolio when the term
structure of interest rates is subject to any kind of variation 215
Overview
We will now generalize the result we obtained in chapter 5, according to
which a xed income investment may well, in certain circumstances, be
immunized against a change in the structure of interest rates. We had sup-
posed that this structure was horizontal and that it would receive a parallel
shift. We are now in a position to suppose that the initial structure has any
kind of shape (not necessarily horizontal); however, we will still suppose
that it undergoes a parallel shift. We will of course have to redene dura-
tion within this somewhat broader framework. For that purpose, we will
use the concept introduced initially by Fisher and Weil.1 The dierence
between their concept and that of Macaulay is that the weights of the times
of payment now make use of the spot rates pertaining to each term, while
Macaulay relied on the bond's yield to maturity to discount the cash ows.
In order to facilitate our calculations, we will work in continuous time.
Let us designate any time structure of spot interest rates as
i(0Y 1)Y i(0Y 2)Y F F F Y i(0Y t)Y F F F Y i(0Y T)
where the rst index (0) designates the present time and the second index
the term we consider. To simplify notation, let i(z) designate the instanta-
neous rate of interest prevailing at time 0 for a loan starting at time z with
1L. Fisher and R. Weil, ``Coping with the Risk of Interest-rate Fluctuations: Return to
Bondholders from Naive and Optimal Strategies.'' The Journal of Business, 44, no. 4 (Octo-
ber 1971).
an innitely small maturity. With this notation, we can easily derive a def-
inite relationship between the spot interest rates i(0Y t)the interest rate
prevailing today for a loan maturing at time tand the implicit instanta-
neous forward rates i(z). We must have the arbitrage-driven relationship
e
_
t
0
i(z) dz
= e
i(0Y t)t
(1)
because $1 invested today at the spot rate i(0Y t) must yield the same
amount as $1 rolled over at all innitesimal forward rates i(z). Conse-
quently, taking the log of (1), we get
i(0Y t) =
_
t
0
i(z) dz
t
(2)
We can see that the spot rate i(0Y t) is none other than the average of all
implicit forward rates. Should the structure of interest rates receive a
shock :, identical for all terms, we can conclude that this would be en-
tirely equivalent to each implicit forward rate receiving the same increase
:. In fact, from (2) we may write
i(0Y t)t =
_
t
0
i(z) dz (3)
This equality remains unchanged if we add :t to each side:
i(0Y t)t :t =
_
t
0
i(z) dz :t
which may also be written as
[i(0Y t) :[t =
_
t
0
i(z) dz
_
t
0
: dz =
_
t
0
[i(z) :[ dz (4)
We can see that increasing each rate i(0Y t) by : is equivalent to displacing
vertically by : the implicit forward rate structure.
We shall denote the present value of a cash ow c(t) received at time t
(per innitesimal period dt)2 as
c(t)e
_
t
0
i(z) dz
2Here we assume a continuously paid coupon; so, for instance, in nominal terms the cash
ow corresponding to the rst year is
_
1
0
c(t) dt =
_
1
0
c dt = c. This means that within any
innitesimal time interval dt, however small, the bond owner receives c dt. Also in nominal
terms, the cash ow for the last year is
_
T
T1
c(t) dt =
_
T
T1
(c B
T
) dt = c B
T
.
210 Chapter 10
or, in an entirely equivalent manner, as
c(t)e
i(0Y t)t
and we shall keep in mind that any increase given to i(z) is equivalent to
an identical increase given to i(0Y t).
Let us now dene a bond's value and its duration when we consider a
term structure of interest rates denoted as i(0Y t).
10.1 Value of a bond in continuous time for a given term structure
Consider a term structure i(0Y t), which we will denote as
~
i for simplicity.
The value of a bond when the structure is
~
i will be equal to
B(
~
i ) =
_
T
0
c(t)e
i(0Y t)t
dt (5)
The reader who is familiar with the calculus of variations will imme-
diately recognize in (5) a functional, that is, a relationship between a func-
tion (
~
i ) and a number (B(
~
i )).
Let us now suppose that the whole structure receives an increase : (in
the language of the calculus of variations, we would say that
~
i receives a
variation :). We consider that this variation takes place immediately after
the bond has been bought. The structure then becomes
~
i :, and the
bond's value is
B(
~
i :) =
_
T
0
c(t)e
[i(0Y t):[t
dt (6)
As a consequence, for any given initial term structure of interest rates, our
bond becomes a function of the single variable :. (This increase in the
function i(0Y t) is merely a particular case of the more general variation
dened by functions such as :(t) in the calculus of variations; here we
have supposed that (t) = 1.)
As the reader may well surmise, duration of the bond with the initial
term structure will be dened as
D(
~
i ) =
1
B(
~
i )
_
T
0
tc(t)e
i(0Y t)t
dt (7)
As before, duration is the weighted average of the bond's times of pay-
ment, the weights being the cash ows in present value. The only dier-
211 Immunizing Against Parallel Moves of Spot Structure
ence is that the cash ows are discounted by a nonconstant structure of
interest rates, and that we are in continuous analysis. We can also see
that, should the structure
~
i receive a variation :, duration will take a new
value, denoted as D(
~
i :), and equal to
D(
~
i :) =
1
B(
~
i :)
_
T
0
tc(t)e
[i(0Y t):[t
dt (7a)
We will soon make use of this last observation.
10.2 A fundamental property of duration
We can show immediately that the logarithmic derivative of the bond's
value with respect to :, with a minus sign, is exactly equal to the bond's
duration. From (6), we can write
1
B(
~
i :)
dB(
~
i :)
d:
=
1
B(
~
i :)
_
T
0
t c(t)e
[i(0Y t):[t
dt = D(
~
i :)
(8)
because of (7a). Also, this relative increase, when : = 0, is equal to the
bond's duration before any change in the structure. Setting : = 0, in (8),
we can write
1
B(
~
i )
dB(
~
i )
d:
=
1
B(
~
i )
_
T
0
tc(t)e
i(0Y t)t
dt = D (9)
because of (7) or (7a).
Consequently, we can see that duration is exactly equal to minus the
relative rate of increase of the bond with respect to a parallel shift in the
structure of interest rates. We will soon make use of this important prop-
erty. (The reader will notice that here, in continuous analysis, duration is
no longer divided by a quantity reminding us of 1 i, which was the co-
ecient by which we had divided duration to get, in discrete time,
1
B
dB
di
;
we called Da(1 i) the modied duration.)
10.3 Value of a bond at horizon H
As we did before, we will now determine the bond's value at horizon H in
order to immunize this future value as well as the rate of return at horizon
212 Chapter 10
H. Let us designate by F
H
(
~
i ) this future value when the term structure of
interest rates is
~
i; we have
F
H
(
~
i ) = B(
~
i ) e
[i(0Y H)[H
(10)
More generally, of course, when the structure becomes
~
i :, we can write
F
H
(
~
i :) = B(
~
i :) e
[i(0Y H):[H
(11)
Our purpose now is to determine a horizon H such that this future
value is equal, at a minimum, to the value it would have if the structure
did not change, that is, if : = 0. This minimal value is the future value of
the bond as calculated today, before the structure undergoes any change;
it is equal to F
H
(
~
i ). We must therefore nd H such that F
H
(
~
i :) goes
through a minimum in space (:Y F
H
), the coordinates of this minimum
being (: = 0Y F
H
= F
H
(
~
i )), as indicated in gure 10.1.
10.4 A rst-order condition for immunization
Minimizing F
H
(
~
i :) is of course equivalent to minimizing ln F
H
(
~
i :);
this is what we will do in order to simplify calculations. With
ln F
H
(
~
i :) = ln B(
~
i :) i(0Y H)H :H (12)
we can write
() (+) 0
F
H
(
i
)
F
H
(
i +
)
F
H
(
i +
)
Figure 10.1
Value of the bond at horizon H, when the term structure of interest rates is
~
i B.
d ln F
H
(
~
i :)
d:
:=0
=
d ln B(
~
i :)
d:
:=0
H = 0 (13)
The last equation implies
H =
1
B(
~
i )
dB(
~
i )
d:
= D(
~
i ) (14)
Our immunizing horizon must be equal to duration, a result that gener-
alizes the one we obtained previously. Of course, the reader will notice
that duration is calculated with respect to the initial structure of interest
rates
~
i (while before duration was calculated simply with respect to the
horizontal structure summarized by the number i ).
10.5 A second-order condition for immunization
We now have to make sure that the function whose derivative we have
just set equal to zero is convex with respect to :, because in such a case we
would have indeed reached a minimum.
Let us calculate the second derivative of ln F
H
(
~
i :). From (12) we
have
d
2
ln F
H
d:
2
=
d
d:
1
B(
~
i :)
dB(
~
i :)
d:
_ _
=
d
d:
[D(
~
i :)[ (15)
(The last equation can be written because of (8).)
The derivative of D(
~
i :) with respect to : is equal to
dD(
~
i :)
d:
=
d
d:
1
B(
~
i :)
_
T
0
tc(t)e
[i(0Y t):[t
dt
_ _
=
1
[B(
~
i :)[
2
__
T
0
t
2
c(t)e
[i(0Y t):[t
dt B(
~
i :)
_
T
0
tc(t)e
[i(0Y t):[t
dt B
/
(
~
i :)
_
=
_
_
T
0
t
2
c(t)e
[i(0Y t):[t
dt
B(
~
i :)
_
T
0
tc(t)e
[i(0Y t):[t
dt
B(
~
i :)
B
/
(
~
i :)
B(
~
i :)
_
(16)
214 Chapter 10
We recognize in the second term of the last brace the expression
D
2
(
~
i :). Simpling the notation, we may thus write
w(t) I
c(t)e
[i(0Y t):[t
B(
~
i :)
Y with
_
T
0
w(t) dt = 1Y
and D(
~
i :) ID(:).
dD
d:
=
_
T
0
t
2
w(t) dt D
2
_ _
=
_
T
0
t
2
w(t) dt 2D
_
T
0
tw(t) dt D
2
_
T
0
w(t) dt
_ _
=
_
T
0
w(t)[t
2
2tD D
2
[ dt =
_
T
0
w(t)[t D[
2
dt (17)
As a happy surprise, the last expression is none other than minus the dis-
persion, or variance, of the terms of the bond,3 and such a variance is of
course always positive. Denoting the bond's variance as S(
~
iY :), we may
write
d
2
ln F
H
d:
2
=
d
d:
[D[ = S(
~
iY :) b 0
Consequently, we may conclude that ln F
H
is indeed convex. Together
with the rst-order condition indicated in section 10.4, this condition is
then sucient for ln F
H
to go through a global minimum at point : = 0
(that is, for the initial term structure of interest rates). Our bond (or bond
portfolio) is thus immunized for a short span of time, whatever parallel
displacement this structure may receive immediately after the bond (or
bond portfolio) has been purchased.
10.6 The diculty of immunizing a bond portfolio when the term structure of
interest rates is subject to any kind of variation
We can now consider immunization in the case of a variable term struc-
ture, when this structure undergoes any kind of variation. We will show
3This result could also have been obtained by observing that the rst part of (17),
_
T
0
t
2
w(t) dt D
2
, itself denes minus the dispersion, or variance, of the terms of the
bond.
that we cannot be sure we can immunize our portfolio, contrary to the
result we obtained previously, when we had considered a parallel shock to
the structure.
Let us dene a variation in the term structure in the following way. Let
i(0Y t) be the initial structure, existing at the time we buy our bond port-
folio. Consider now (0Y t), another term structure of interest rates, and
:a number that, as before, may be positive, negative, or equal to zero.
We will call :(0Y t) a variation of the structure; this variation is supposed
to occur immediately after the purchase of the bond portfolio. The new
structure is thus i(0Y t) :(0Y t). Recapitulating, we have
.
Initial term structure of interest rates: i(0Y t)
.
Variation of the structure: :(0Y t)
.
New structure: i(0Y t) :(0Y t)
Figure 10.2 illustrates an example of each of these functions.
To each structure i(0Y t) :(0Y t) corresponds a value of the bond (or
of the portfolio) equal to
Maturity
t T
Spot rate
structure
i (0, t) + (0, t)
new structure
Variation of the structure
(0, t)
Initial structure
i (0, t)
Figure 10.2
Initial term structure of interest rates i(0, t), variation of the structure h(0, t), and new struc-
ture i(0, t)Bh(0, t).
216 Chapter 10
B[i(0Y t) :(0Y t)[ =
_
T
0
c(t)e
[i(0Y t):(0Y t)[t
dt (18)
Clearly, the bond's value is a functional, that is, a relationship between
a function of tin our case the whole structure i(0Y t) :(0Y t)and a
number (the bond's value B). But the functions i(0Y t) and (0Y t) can be
considered as xed. (At the time of the bond's purchase, i(0Y t) is xed;
immediately after this purchase, the structure i(0Y t) undergoes a variation
:(0Y t), where (0Y t) can be considered as xed, although the investor
was unaware of this at the time he bought the bond.) As a consequence, B
depends on : only. We have thus transformed our functional into a
function of the sole variable :. We will therefore denote this function as
B(:).
By taking the derivative of B(:), we can see that duration is no longer
equal to the logarithmic derivative of B with respect to :, with a minus
sign. Indeed, we can calculate that
1
B
dB
d:
:=0
=
1
B
_
T
0
t(0Y t)c(t)e
i(0Y t)t
dt
This expression is not equal to the duration before the shock to the
structure, because the expression of duration would be the right-hand side
of the above equation without the variation (0Y t), which was unknown
at the time of the bond's purchase.
The value of the bond at horizon H, after the shock to the structure has
happened, is equal to
F
H
= B(:)e
[i(0Y H):(0Y H)[H
(19)
In order to minimize this value with respect to :, it would be sucient,
as before, to minimize ln F
H
:
ln F
H
= ln B(:) [i(0Y H) :(0Y H)[H (20)
Taking the derivative of (20) with respect to :, we get
d ln F
H
d:
=
d ln B(:)
d:
(0Y H)H
Unfortunately, it is not possible in this case to choose a horizon H
+
that
would minimize F
H
; indeed, we would have to be able to choose H
+
such
that
H
+
=
d ln B(:)
d:
(0Y H)
:=0
and this is impossible since we do not know the variation (0Y t), although
this was equal to 1 in the preceding case of a parallel displacement of the
term structure. Hence it is generally not possible to immunize our bond
portfolio if the structure is subject to any kind of variation.
There remains the question regarding the kind of immunization process
we want to undertake: shall we aim at protecting our investor against a
parallel shift in the term structure, supposing that initially it is at or that
it has an initial nonat shape? In the rst case we would consider the
Macaulay duration, and in the second we would have to calculate the
Fisher-Weil duration, which implies calculating rst the term structure.
Fortunately, many tests have been performed, and the second strategy
does not seem to yield signicantly better results than the rst one. Pos-
sibly, the reason for this is twofold: rst, the term structure is relatively
at where it counts most (that is, for long maturities); and second, it is
quite possible that in the long run the nonparallel displacements of the
term structure have a tendency to compensate each other in their eects,
since over long periods the shape of the structure generally increases
smoothly, as we have explained in this text.
A hopefully satisfactory answer to immunization of any kind of change
in the spot rate structure will have to wait until chapter 15. There we will
propose a new method of immunizing a portfolio and will give quite a
number of applications, where we will consider dramatic changes in the
spot rate structure.
Main formulas
.
Value of a bond in continuous time, with
~
i Ii(0Y t) being the term
structure or interest rates:
B(
~
i ) =
_
T
0
c(t)e
i(0Y t)t
dt (5)
.
Duration of the bond:
D(
~
i ) =
1
B(
~
i )
_
T
0
tc(t)e
i(0Y t)t
dt (7)
218 Chapter 10
.
Duration of the bond when
~
i receives a (constant) variation ::
D(
~
i :) =
1
B(
~
i :)
_
T
0
tc(t)e
[i(0Y t):[t
dt (7a)
.
Fundamental property of duration:
1
B(
~
i )
dB(
~
i )
d:
=
1
B(
~
i )
_
T
0
tc(t)e
i(0Y t)t
dt = D (9)
.
First-order condition for immunization:
H =
1
B(
~
i )
dB(
~
i )
d:
= D(
~
i ) (14)
.
Second-order condition for immunization:
d
2
ln F
H
d:
2
b 0 =
d
d:
[D(
~
i :)[ = S(
~
iY :) b 0
Questions
10.1. How do you derive equation (8) in section 10.2 of this chapter?
10.2. This question could be considered as a project. You have noticed
that in section 10.3, we sought to protect the investor against a parallel
displacement of the term structure for interest rates by seeking a horizon
such that the future value of the bond (or the bond portfolio), F
H
, goes
through a minimum for the initial term structure, that is, for : = 0 in
(F
H
Y :) space. How would you have obtained the ensuing immunization
theorem if you had considered nding a minimum for the investor's ho-
rizon rate of return? (Hint: First dene carefully the investor's horizon
rate of return by extending the concept you encountered in section 6.2 of
Chapter 6 to the concepts introduced in this chapter.)
11
CONTINUOUS SPOT AND FORWARD
RATES OF RETURN, WITH TWO
IMPORTANT APPLICATIONS
Hotspur. Are there not some of them set forward already?
Henry IV
Chapter outline
11.1 The continuously compounded rate of return 222
11.2 Continuously compounded yield 226
11.3 Forward rates for instantaneous lending and borrowing, and spot
rates 228
11.4 Relationships between the forward rates, the spot rates, and a
zero-coupon bond price 232
Overview
Until now we have calculated rates of return over nite periods of time
(for instance, one year). There are, however, considerable advantages to
considering rates of return over innitesimally small amounts of time. A
rst reward is to simplify many computations. For example, the reader
will recall that, in chapter 7, with numbers of years n
1
` n
2
` n
3
, the n
2
spot rate was a weighted geometric mean of 1 plus the n
1
spot rate and 1
plus the n
3
spot rate minus 1. This is a rather cumbersome expression,
whose properties are not easy to analyze. When considering continuously
compounded rates, we will be able to replace ane transformations of
geometric averages with simple arithmetic averages. But by far the most
important advantage of such a procedure is that it enables us to apply the
central limit theorem and thus derive from that major result of statistics a
theory about the probabilistic nature of both dollar annual returns and
asset prices. We will thus be able to justify the lognormal distribution of
the dollar returns on assets, as well as the lognormal distribution of the
asset prices.
This chapter will be organized as follows: we will rst introduce the
concept of a continuously compounded rate of return on an arbitrary
time interval. We will then move on to describe in the same context yields,
spot rates, and forward rates, and their relationships to zero-coupon bond
prices. We will nally apply the central limit theorem in order to have
a theory about the probability distribution of dollar returns and asset
prices.
11.1 The continuously compounded rate of return
Consider, on the time axis, a length of time [uY v[ of size v u and an
arbitrary partition of this length into n intervals denoted hz
1
, hz
2
, F F F , hz
n
.
These intervals, not necessarily of the same length, are measured in years;
their sum,
n
j=1
hz
j
, is equal to v u. Consider, without any loss in gen-
erality, the rst interval, hz
1
. Suppose that we are at time u (at the begin-
ning of this interval) and that we have agreed to lend capital C
u
at that time
at annual rate i
1
during the amount of time hz
1
. The contract species that
the interest will be computed and paid at the end of that period (at time
u hz
1
) and will be in the amount of i
1
hz
1
. Indeed, hz
1
is simply the
fraction of year corresponding to this contract. For our purposes, we could
consider that hz
1
is shorter than one year. In order to stress the fact that the
contract stipulates that the interest is paid once per period, we write it i
(1)
1
,
where the superscript denotes the number of times the annual rate of
interest is paid within hz
1
; C
u
i
(1)
1
hz
1
is then the interest paid at u hz
1
.
Including principal, you would receive C
u
C
u
i
(1)
1
hz
1
= C
u
(1 i
(1)
1
hz
1
).
Consider now another contract. This time, the contract stipulates that
the interest will be paid twice per period hz
1
. The amount received
as interest in the middle of period hz
1
, at time u
1
2
hz
1
, would be
C
u
i
(2)
1
hz
1
2
; the total amount received at time u hz
1
would therefore be
C
u
_
1
i
(2)
1
2
hz
1
_
2
.
The generalization to a contract paying interest m times, at equal dis-
tances within the interval hz
1
, is immediate. It would pay the amount
C
uhz
1
= C
u
1
i
(m)
1
m
hz
1
_ _
m
(1)
at time u hz
1
. Of course, this amount is dierent from the amount
corresponding to a contract with an interest rate i
(1)
1
paid once per period.
In fact, we can show that it is always superior to it; with i
(1)
1
= i
(m)
1
,
1 i
(1)
1
hz
1
is simply the sum of the rst two terms of the binomial devel-
opment of (1 i
(m)
1
hz
1
am)
m
and is therefore always strictly inferior to
u v
Time
hz
1
hz
2
hz
j
hz
n
222 Chapter 11
it. Thus, the i
(m)
1
contract is denitely better for the lenderand more
expensive for the borrowerthan the i
(1)
1
contract. For both contracts to
produce the same amount, i
(m)
1
should be related to i
(1)
1
by the following
relationship:
1
i
(m)
1
m
hz
1
_ _
m
= 1 i
(1)
1
hz
1
(2)
and therefore i
(m)
1
should be set equal to
i
(m)
1
=
m
hz
1
[(1 i
(1)
1
hz
1
)
1am
1[ (3)
For example, suppose i
(1)
1
= 8% per year; hz
1
= 3 months = 0X25 year;
m = 10. From (3), i
(m)
1
should be equal to 7.9289% per year for both
contracts to be equivalent, both producing 1.02 C
u
at time u hz
1
.
Consider now what happens when m, the number of times the interest is
paid throughout the interval hz
1
, increases to innity. Setting in (1)
i
(m)
1
hz
1
m
=
1
k
, with m = ki
(m)
1
hz
1
, we have
C
uhz
1
= C
u
1
1
k
_ _
ki
(m)
1
hz
1
(4)
In the limit when m y we have also k yand therefore
lim
ky
C
uhz
1
= C
u
e
i
(y)
1
hz
1
(5)
an all-important relationship, where i
(y)
1
is called the continuously com-
pounded rate of interest prevailing throughout interval hz
1
. For the i
(y)
1
contract to yield the same amount as the i
(1)
1
contract, we should have
e
i
(y)
1
hz
1
= 1 i
(1)
1
hz
1
(6)
or, equivalently,
i
(y)
1
=
log(1 i
(1)
1
hz
1
)
hz
1
(7)
Coming back to our former example, i
(y)
1
should be equal to
log(10X02)
0X25
=
7X921% for both contracts to yield 1X02C
u
at time u hz
1
, that is, after
three months.
Notice an important special case for (7): if hz
1
= one year, the equiva-
lence between the continuously compounded interest rate and the once-
223 Continuous Spot and Forward Rates of Return
per-year compounded (or simple) interest rate is, suppressing the ``1''
subscript for simplicity,
i
(y)
= log(1 i
(1)
) (8)
Is the continuously compounded annual interest rate i
(y)
= log(1 i
(1)
)
close in some sense to the simple interest rate i
(1)
? The answer is yes. It
turns out that i
(1)
is nothing else than the linear approximation of i
(y)
=
log(1 i
(1)
). Indeed, the rst-order (or linear) Taylor development of
log x at x = 1 is x 1; thus the same development of log(1 i
(1)
1
) at
i
(1)
1
= 0 is simply i
(1)
. (From a geometric point of view, the ray i
1
is tan-
gent to the curve i
(y)
= log(1 i
(1)
) at point i
(1)
= 0.) Thus, the smaller
the value of i
(1)
, the better the approximation and the closer the con-
tinuously compounded rate i
(y)
is to the simple rate i
(1)
. For instance,
table 11.1 gives i
(y)
as a function of i
(1)
and illustrates the convergence of
these rates when i
(1)
0.
Figure 11.1 represents the continuously compounded interest rate i
(y)
as a function of the simple rate. The ``straight line'' character of i
(y)
is a
surprise. It comes from the fact that even when i
(1)
=15%, we are still very
close to zero and therefore the linear approximation is still very good.
We now point to a nal property of the continuously compounded
interest rate: that rate is the only rate that makes an investment's value
come back to its initial value when the rate has had a given value in one
period (one year, for instance) and the opposite value in a second period
of equal length. Indeed, consider three consecutive years, j 1, j, and
j 1, and let i
(y)
denote for simplicity the continuously compounded
interest rate; we have
S
j1
= S
j1
e
i
(y)
e
i
(y)
= S
j1
Table 11.1
Convergence of the equivalent continuously compounded interest rate i
(L)
toward the simple
interest rate i
(1)
when i
(1)
tends toward zero (in percentage per year)
Simple interest rate
i
(1)
Equivalent continuously
compounded interest rate
i
(y)
= log(1 i
(1)
)
Dierence
i
(1)
i
(y)
0 0 0
2 1.98 0.02
4 3.92 0.08
6 5.83 0.17
8 7.70 0.30
10 9.53 0.47
224 Chapter 11
Therefore, S does not change value. This would not be true with an
interest rate compounded a number m (1 m ` y) times per period; if
i
(m)
denotes such a rate of return, a negative dierence between S
j1
and
S
j1
would always appear:
S
j1
= S
j1
1
i
(m)
m
_ _
m
1
i
(m)
m
_ _
m
= S
j1
1
i
(m)
m
_ _
2
_ _
m
The coecient multiplying S
j1
is clearly smaller than 1, and therefore
S
j1
` S
j1
. This inequality becomes an equality if and only if m tends
toward innity, in other words, if and only if the rate of return is com-
pounded an innite number of times per period. Indeed, we then get
lim
my
S
j1
= lim
my
S
j1
1
i
(m)
m
_ _
m
1
i
(m)
m
_ _
m
= S
j1
e
i
(y)
e
i
(y)
= S
j1
(9)
Let us now come back to our partition of the time span [uY v[. Consider
that contracts always provide for continuously compounded interest rates
i
(y)
j
, j = 1, F F F , n.
0 5 10 15 20
0
20
15
10
5
Simple interest rate (in %/year)
Simple rate of interest: i
(1)
Continuously compounded
rate of interest: i
()
= log (1 + i
(1)
)
S
i
m
p
l
e

a
n
d

c
o
n
t
i
n
u
o
u
s
l
y

c
o
m
p
o
u
n
d
e
d
i
n
t
e
r
e
s
t

r
a
t
e
s
Figure 11.1
The proximity of the equivalent continuously compounded interest rate and the simple interest
rate. The latter is the linear approximation of the former at i
(1)
= 0.
At time u hz
1
, C
u
has become C
u
e
i
(y)
1
hz
1
. By the same reasoning,
at time u hz
1
hz
2
(at the end of interval hz
2
), C
u
becomes
C
u
e
i
(y)
1
hz
1
Xe
i
(y)
2
hz
2
= C
u
e
i
(y)
1
hz
1
i
(y)
2
hz
2
, and of course at time v (at the end of
the last interval hz
n
) C
u
becomes
C
v
= C
u
e
n
j=1
i
(y)
j
hz
j
(10)
Until now the rate of interest was considered a discontinuous function
of time. Indeed, since the i
(y)
j
are not necessarily equal, i
(y)
j
is a function
of time dened by a succession of constant values over each interval hz
j
.
Since these values are not necessarily equal to each other, we may have a
series of n 1 successive jumps for the interest rate between time u and
time v. Consider now what happens to our expression (10) if i
(y)
j
becomes
a continuous function of time, dened over innitesimally small time
intervals. (Note that we could still consider a discontinuous function, but
it is not necessary to do so for our purposes here.) To that eect, trans-
form the partition of [uY v[ in such a way that the number of intervals n
tends toward y, and the largest interval hz
j
tends toward zero. If for any
partition the sum
n
j=1
i
(y)
j
hz
j
tends toward a limit, that limit is the de-
nite integral of i
(y)
j
between u and v. We thus have
lim
ny
max hz
j
0
n
j=1
i
(y)
j
hz
j
=
_
v
u
i(z) dz (11)
where i(z) now denotes the continuously compounded interest rate as a
function of time, and therefore C
v
can be written as
C
v
= C
u
e
r
v
u
i(z) dz
(12)
11.2 Continuously compounded yield
Consider an investment C
u
that is transformed into C
v
after a time span
v u. We dene the yearly continuously compounded yield as the (con-
226 Chapter 11
stant) continuously compounded interest rate R(uY v) that transforms C
u
into C
v
in time v u. Therefore, applying what we have done before at
the level of any time interval, we can dene R(uY v) implicitly as
C
u
e
R(uY v)(vu)
= C
v
(13)
and thus R(uY v) is equal to
R(uY v) =
log(C
v
aC
u
)
v u
(14)
There is an important relationship between the continuously com-
pounded return and the yield. From (12) and (13) we may write
C
u
e
r
v
u
i(z) dz
= C
u
e
R(uY v)(vu)
(15)
implying
R(uY v) =
_
v
u
i(z) dz
v u
(16)
and therefore the continuously compounded yield is equal to the average
value of the continuously compounded interest rate. These relations provide
a nice interpretation of the Euler number. We use equations (13) and (16)
to write
C
v
= C
u
exp
r
v
u
i(z) dz
v u
(v u)
_ _
= C
u
e
R(uY v)(vu)
(17)
Consider an investment of C
u
= $1 and a yield over [uY v[ equal to
1
vu
.
Then from (17), C
v
= e = 2X71828 F F F X We can state the following:
The Euler number e is what $1 becomes after a time span v u
years when the continuously compounded yield (the average of the
continuously compounded instantaneous interest rates) is
1
vu
.
Indeed, we would have C
v
= 1 e
1
vu
(vu)
= e. For instance, suppose that
v u is 20 years; then, with R(uY v) = 1a(v u) = 5%/year, $1 becomes
$2.71828 after 20 years. Of course, this holds for fractional years just as
well: if (v u) = 20X202 years and R(uY v) =
1
20X202 years
= 4X95%/year, $1
becomes e dollars 20.202 years later.
Notice the generality of this interpretation. It does not suppose that all
interest rates are necessarily positive; it would apply to the real value of $1
at time v if, during some subintervals of [uY v[, real interest rates were
negative, that is, nominal interest rates were lower than instantaneous in-
ation rates.
11.3 Forward rates for instantaneous lending and borrowing, and spot rates
Until now, we have supposed that when n was nite, there was a nite
number of contracts that were signed at the various dates u, u hz
1
,
u hz
1
hz
2
, . . . , that is, at the beginning of each interval hz
j
. But we
could of course simplify our scenario and imagine that at date u there
exist n forward markets, each with a time span hz
j
, and that one can enter
all those contracts at initial time u. Therefore, our interest rates pertaining
to any interval j, denoted i
j
, would be replaced by forward rates. Now
consider what happens when the number of periods goes to innity and
the duration of the loan goes to zero. We can dene an instantaneously
compounded forward rate for innitely short borrowing as f (uY z), where
u is the contract inception time and z (z b u) is the time at which the loan
is made, for an innitesimally small period. We thus have
f (u, z) = instantaneously compounded forward rate for
a loan contracted at time u, starting at time z,
for an infinitesimal period
Let us now dene the spot rate at time u as the continuously com-
pounded return that transforms C
u
into C
v
after a time span v u.
Applying what we saw in section 11.2, this is exactly the continuously
compounded yield, which we can therefore denote as R(uY v), such that
C
v
= C
u
e
R(uY v)(vu)
. If agents have costless access to all forward markets in
their capacity both as lenders and as borrowers, arbitrage ensures that $1
invested in all forward markets between u and v will be transformed into
the same amount as $1 invested at the continuously compounded spot
rate R(uY v). We must have
e
r
v
u
f (uY z) dz
= e
R(u)du
(18)
228 Chapter 11
and therefore
R(uY v) =
_
v
u
f (uY z) dz
v u
(19)
Equation (19) is important. It tells us that the spot rate R(uY v) is equal to
the arithmetic average of the forward rates f (uY z), z e [uY v[. Note that
this relationship, valid when all rates are continuously compounded, is
much simpler than the relationship between the spot rate and the forward
rates when these are compounded a nite number of times. (Indeed, we
showed that when the compounding took place once a year, one plus the
spot rate was a geometric average of one plus the forward ratesa much
more cumbersome relationship.) Using the arithmetic average property
expressed in (19), or the arbitrage equation (18), we can write
_
v
u
f (uY z) dz = R(uY v)(v u) (20)
and take the derivative of (20) with respect to v to obtain a direct rela-
tionship between any forward rate and the spot rate:
f (uY v) = R(uY v) (v u)
dR(uY v)
dv
(21)
Equation (21) has a nice economic interpretation. Suppose that the spot
rate R(uY v) is increasing. This means that the average rate of return
increases. What does this increase imply in terms of the forward (the
``marginal'') rate? In other words, given an increase in the spot rate, what
is the implied value of the forward rate in relation to both the value of the
initial spot rate and the increase this spot rate has received? The answer is
quite simple and follows the rules relating average values to marginal
values: the forward rate (the marginal rate of return) must be equal to the
spot rate (the average rate of return) plus the rate of increase of the spot
rate multiplied by the sum of all innitesimally small time increases be-
tween u and v (this sum is equal to (v u)). Thus the forward rate must
be equal to R(uY v) (v u)
dR(uYv)
dv
as indicated in (21). Of course, the same
reasoning would have applied had we considered a decreasing spot rate.
The relationship between the forward rate and the spot rate, as
expressed in (21) or in its integral forms ((19) or (20)), entails useful
properties of the forward rate. We can see immediately that:
1. the forward rate is higher than the spot rate if and only if the spot rate
is increasing;
2. the forward rate is lower than the spot rate if and only if the spot rate
is decreasing; and
3. the forward rate is equal to the spot rate if and only if the spot rate is
constant.
One caveat is in order here: an increasing spot rate curve does not
necessarily entail an increasing forward rate curve. Indeed, an increasing
average implies only a marginal value that is above it (and not necessarily
an increasing marginal value). In fact, the forward rate can be increasing,
decreasing, or constant, while the spot rate is always increasing. To illus-
trate this, consider the well-known, common case where the spot rate
slowly increases and then reaches a constant plateau. We can show that in
this case if the spot rate is a concave function of maturity, the forward
rate will always be decreasing in an interval to the left of the point where
the spot rate reaches its maximum.
Let [0Y T[ denote our interval [uY v[. We then write
R(0Y T) =
_
T
0
f (0Y z) dz
T
(22)
Suppose that R(0Y T) is an increasing, concave function of T, reaching a
maximum at

T with a zero slope at

T. This translates as
dR(0Y

T)
dT
= 0 (23)
and
d
2
R(0Y

T)
dT
2
` 0 (24)
Therefore, (23) implies, from (21),
f (0Y

T) = R(0Y

T) (25)
230 Chapter 11
On the other hand, (24) and (25) imply
d
2
R(0Y

T)
dT
2
=
1
T
df (0Y

T)
dT

dR(0Y

T)
dT
_ _
` 0
and therefore, since
dR(0Y

T)
dT
= 0,
d f (0Y

T)
dT
` 0. Thus the forward rate must be
decreasing when the spot rate reaches its plateau.
The following example illustrates what we have just shown. Suppose
that the spot rate R(0Y T), expressed in percent per year, is the following
function of maturity:
R(0Y T) =
0X005T
2
0X2T 4 (for maturities 0 to 20 years)
6 (for maturities beyond 20 years)
_
Applying the properties of the forward rate, we know that it will be
equal to the spot rate (6% per year) for maturities beyond 20 years. To
recover the forward rate for maturities between 0 and 20 years, we could
apply (21), replacing u and v by 0 and T, respectively. But perhaps a more
elegant way to go about it would be to come back to the arbitrage rela-
0 5 10 15 20 25 30
0
8
7
6
5
4
3
Maturity T
Forward rate f (0, T) = 0.015T
2
+ 0.4T + 4
f (0, T) = R(0, T) = 6%
S
p
o
t

r
a
t
e

a
n
d

f
o
r
w
a
r
d

r
a
t
e

(
i
n

%
/
y
e
a
r
)
Area
0
f (0, z)dz
T
Spot rate R(0, T) = 0.005T
2
+ 0.2T + 4
Figure 11.2
An example of the correspondence between the forward rate f (0, T) for instantaneous bor-
rowing and the spot rate R(0, T). R(0, T) is the average of all forward rates between 0 and T.
Hence area R(0, T )
.
TF f
T
0
f (0, z) dz.
tionship between spot and forward rates, which implies
R(0Y T) T =
_
T
0
f (0Y z) dz (26)
In our example, this yields
0X005T
3
0X2T
2
4T =
_
T
0
f (0Y z) dz (27)
Taking the derivative of (27) with respect to T yields the forward curve
between T = 0 and T = 20:
f (0Y T) = 0X015T
2
0X4T 4
Both the spot rate curve and the forward rate curve are shown in gure
11.2. Indeed, it can be veried that there is a whole range of maturities
where the forward curve is decreasing while the spot rate is increasing;
that range is between maturities 13X
"
3 years (corresponding to the maxi-
mum of the forward rate curve) and 20 years.
11.4 Relationships between the forward rates, the spot rates, and a zero-coupon
bond price
Consider a defaultless zero-coupon bond at time t, which matures at time
T with $1 face value. Denote the price of that bond as B(tY T). Suppose
also that today is time 0; so, with t b 0, B(tY T) is a random variable.
Nevertheless, it can be related in a very precise way to the future forward
rates f (tY u) that will apply at time t and to the future spot rate R(tY T)
that will apply at time t. Applying the denition of forward rates, we can
rst write
B(tY T)e
r
T
t
f (tY u) du
= 1 (28)
from which we deduce
B(tY T) = e
r
T
t
f (tY u) du
(29)
In (29), the bond's value B(tY T), a random value, is a function of all
values f (tY u) that the forward rate will take at time t for horizons from
u = t to u = T. For each horizon u (now xing u), the value f (tY u) may
232 Chapter 11
be considered a random process starting at a known value (today) f (0Y u)
and having properties that we will describe in later chapters. For the time
being, it is important to note that the zero-coupon bond's value B(tY T) is
a function of an innitely large number of outcomes of random variables
(the forward rates f (tY u), where u ranges between t and T), each of them
being the outcome of a random process taking place between 0 and t.
Alternatively, we could express the future bond's value in terms of the
future spot rate R(tY T). We can write
B(tY T)e
R(tY T)(Tt)
= 1 (30)
and, therefore,
B(tY T) = e
R(tY T)(Tt)
(31)
This time B(tY T) is a function of the single random variable R(tY T),
which may also be considered the result of a random process started at
time 0. In any case, we can see that we could model as well the evolution
of each forward rate f (tY u) (for each u) or the evolution of the single spot
rate R(tY T) as a random process starting at t. Note that a remarkably
comprehensive modelthat of Heath, Jarrow, and Morton (1992)
follows the rst route and models the forward rates. (This is the model we
shall describe and use from chapter 17 onward.)
Main formulas
.
Equivalence relationship between the once-per-year compounded (or
simple) interest rate i
(1)
and the continuously compounded interest rate
i
(y)
:
i
(y)
= log(1 i
(1)
) (8)
or
i
(1)
= e
i(y)
1
.
Over a time span [uY v[ partitioned into n intervals hz
j
, j = 1, F F F , n, a
sum C
u
becomes
C
v
= C
u
e
n
j=1
i
(y)
j
hz
j
(10)
.
If the following limit exists:
lim
ny
max hz
j
0
n
j=1
i
(y)
j
hz
j
=
_
v
u
i(z) dz (11)
C
v
= C
u
e
r
v
u
i(z) dz
(12)
.
Continuously compounded yield = continuously compounded spot
rate = R(uY v) such that
C
u
e
R(uY v)(vu)
= C
v
(13)
or
R(uY v) =
log(C
v
aC
u
)
v u
(14)
.
Relationship between continuously compounded rate of return over
period [uY v[, R(uY v), and instantaneous rate of interest i(z):
R(uY v) =
_
v
u
i(z) dz
v u
(16)
The continuously compounded rate of return is the average of the
instantaneous rates of interest i(z) over period [uY v[.
.
Economic interpretation of the Euler number:
The Euler number e is what $1 becomes after a time span v u years
when the continuously compounded yield (the average of the con-
tinuously compounded instantaneous interest rates) is
1
vu
.
.
Denition: f (uY z) = instantaneously compounded forward rate for a
loan contracted at time u, starting at time z, for an innitesimal period.
.
Relationship between forward rates f (uY z) and continuously com-
pounded spot rate R(uY v):
R(uY v) =
_
v
u
f (uY z) dz
v u
(19)
equivalent to
_
v
u
f (uY z) dz = R(uY v)(v u) (20)
234 Chapter 11
and to
f (uY v) = R(uY v) (v u)
dR(uY v)
dv
(21)
Questions
11.1. Consider a contract over a period of six months, with a yearly rate
of interest of 7%. Suppose that the interest is paid only once at the end of
that period. What should the rate of interest be for a contract yielding the
same return after six months if the interest is to be paid after each month?
11.2. Using the same data as in question 11.1, what should the con-
tinuously compounded rate of interest be for the contracts to yield the
same return?
11.3. In section 11.3 we showed that the continuously compounded spot
rate over a period v u (R(uY v), or R(0Y T) if we consider a time interval
T) was the average of all forward rates for innitesimal trading periods
over this interval (see equation (19)). This in turn implies that the product
of the time interval (v u, or T) by the spot rate is equal to the sum (the
integral in this case) of all forward rates from u to v (or from 0 to T), such
that
_
v
u
f (uY z) dz or
_
T
0
f (0Y z) dzsee equation (20). The aim of this
exercise is to verify this result by considering the data in gure 11.2.
a. Consider, as we have done, a time interval of 10 years. First, verify
that the spot rate R(0Y 10) is 5.5% per year according to the hypothesis we
have chosen for the interest rate structure.
b. Verify that the product of this spot rate by the time interval (the area
of the rectangle with height equal to the spot rate and width equal to the
time interval) is indeed equal to the denite integral of the forward rate
between 0 and 10.
c. What are the measurement units of this rectangle?
d. Give an economic, or nancial, interpretation of the area of this rect-
angle or of the denite integral whose value is equal to that area.
e. Does your answer to (d) lead you to think that you could have dis-
pensed with performing the calculation of the denite integral corre-
sponding to (b)?
12
TWO IMPORTANT APPLICATIONS
Maria. . . . all the power thereof it doth apply . . .
Love's Labour's Lost
Chapter outline
12.1 A theoretical justication of the lognormal distribution of asset
prices 238
12.2 In search of the term structure of interest rates, or to spline or not
to splineand how? 240
12.2.1 Parametric methods 244
12.2.1.1 The Nelson-Siegel model 244
12.2.1.2 The Svensson model 247
12.2.2 Spline methods: the Fisher-Nychka-Zervos, the
Waggoner, and the Anderson-Sleath models 248
12.2.2.1 What is a spline? 248
12.2.2.2 Some strangeand niceconsequences 249
12.2.2.3 Determining the order of the optimal
polynomial spline 251
12.2.2.4 The Fisher-Nychka-Zervos model 253
12.2.2.5 An extension of the Fisher-Nychka-Zervos
model: the Waggoner model 253
12.2.2.6 The Anderson-Sleath model 256
Appendix 12.A.1 A proof of the central limit theorem and a
determination of the probability distribution of the dollar return on
an asset 257
Appendix 12.A.2 Deriving the Euler-Poisson equation from economic
reasoning 259
Overview
We will now apply the concepts of continuous spot and forward rates
to two central subjects of nance. The rst is the theoretical justication
of the lognormal distribution of asset prices. It is a hypothesis that has
been at the core of the evaluation of assets and their derivatives for the
past several decades. The second is the search for the term structure of
interest rates, an extraordinarily elusive object but a highly interesting one.
Indeed, for the last 30 years or so, researchers have displayed tremendous
ingenuity in attempting to extract the term structure of interest rates from
bond prices. We will describe recent advances in this eld.
12.1 A theoretical justication of the lognormal distribution of asset prices
The hypothesis of lognormally distributed asset prices is among the most
widespread assumptions in nancial economics. We will now show how it
can be justied through the central limit theorem; we will also show that
alternatively, it can be founded on a Wiener process.
Let a unit time span (one year, for instance) be divided into n equal
intervals (one interval is, for instance, one day). Let j denote any of these
intervals; j = 1, . . . , n. By convention j will indicate the end of the time
interval, and j 1 its beginning. An asset is priced at S
j1
at the begin-
ning of the j th interval and at S
j
at its end. Let r
j1; j
denote the daily
continuously compounded rate of return on S within one interval, thus
equal to log(S
j
=S
j1
).
Let us now express S
n
, the asset's value after n periods, equivalently,
after one year, as follows:
S
n
= S
0

S
1
S
0
. . .
S
j
S
j1
. . .
S
n
S
n1
= S
0
e
r
0; 1
. . . e
r
j1; j
. . . e
r
n1; n
= S
0
e
n
T
j=1
r
j1; j
(1)
Thus S
n
is expressed as a simple exponential whose exponent is a sum of
the n random variables r
j1; j
.
Suppose now that these random variables are independent, identically
distributed with mean m and (nite) variance s
2
. We can now appeal to
the central limit theorem: the sum
n
j=1
r
j1; j
will be normally distributed
with mean
n
j=1
m = nm and variance
n
j=1
s
2
= ns
2
when n is large. (A
proof of the theorem can be found in appendix 12.A.1). Call the mean
and variance m and s
2
, respectively; let
n
j=1
r
j1; j
1r
0; n
. We then have
S
1
= S
0
e
r
0; 1
; r
0; 1
@N(m; s
2
) (2)
if time is expressed in years; thus S
1
=S
0
is a lognormal variable with
parameters (m; s
2
).
238 Chapter 12
It can be seen that the lognormal property of an asset's value is inde-
pendent from the peculiar distribution of the continuously compounded
rate of return of the asset over an interval, so long as the conditions of the
central limit theorem are met. The strength of the theorem stems precisely
from the fact that these conditions are quite unrestrictive: the con-
tinuously compounded daily rates of return, r
j1; j
, are simply supposed to
be independent and identically distributed. The remarkable property of
the lognormally distributed variable S
1
=S
0
stems from the fact that it will
apply whatever the type of distribution followed by the r
j1; j
variable is,
so long as it has nite variance and the r
j1; j
's are independent.
There is another way of justifying directly the lognormality of asset
prices, which rests on the assumption that the continuously compounded
rate of return between any time 0 and time t, log(S
t
=S
0
), follows a general
Wiener process. Dene W
t
as a Wiener process such that W
0
= 0,
W
t
@N(0; t), with t
1
< t
2
, DW
t
= W
t
2
W
t
1
@N(0; t
2
t
1
) 1N(0; Dt).
Suppose that
log(S
t
=S
0
) = mt sW
t
= mt s
t
_
e
t
; e
t
@N(0; 1) (3)
or
log(S
t
=S
0
) @N(mt; s
2
t)
Over a nite interval Dt, the continuously compounded rate of return,
log S
tDt
=S
t
, is
log
S
tDt
S
t
_ _
= log
S
tDt
S
0
_ _
log
S
t
S
0
_ _
= m (t Dt) sW
tDt
mt sW
t
= mDt s(W
tDt
W
t
) = mDt sDW
t
= mDt s
Dt
_
e
t
; e
t
@N(0; 1)
Hence, S
tDt
=S
t
is also lognormal and can be written
S
tDt
S
t
= e
mDts
Dt
_
e
t
; e
t
@N(0; 1)
239 Two Important Applications
Two special cases are important: Dt = 1 year and Dt = p years. We have,
respectively,
S
1
S
0
= e
mse
t
which implies log(S
1
=S
0
) = m se
t
@N(m; s
2
) as before, and
S
p
S
0
= e
mps

p
_
e
t
entailing log(S
p
=S
0
) @N(mp; s
2
p).
Thus the same results are obtained using two dierent probabilistic
assumptions. The rst is that over very short intervals the continuously
compounded rates of return are independent and identically distributed,
and we are then free to apply the central limit theorem. The second is a bit
more restrictive because it assumes that the continuously compounded
return over the same small interval follows a process with very precise
propertiesthe Wiener process.
12.2 In search of the term structure of interest rates, or to spline or not to spline
and how?
Before dealing with the important ow of current research in this area, let
us remember how the following three concepts are dened and related to
each other (in continuous time):
.
the price today of a series of default-free zero-coupon bonds with
various maturities,
.
the spot rate, or zero-coupon yield curve, and
.
the forward rate curve.
We will take advantage of this quick refresher to simplify the notation
somewhat in order to make it identical or immediately comparable to that
used in recent research on the subject.
First, in section 11.4 we denoted the price at time t of a default-free
zero-coupon bond paying one monetary unit at maturity T as B(t; T). We
will now simplify this notation by supposing that time today is 0. So
one zero-coupon bond price can be written as B(0; T), or, more com-
240 Chapter 12
pactly, as B(T), omitting the rst index. This function of T is often called
the discount function.
The spot rate is the yield to maturity of that zero-coupon bond, denoted
R(0; T) in section 11.4. We can now do away with the rst index and
write simply R(T). It is equal to the yield-to-maturity of the zero-coupon.
Hence its relationship to the zero-coupon's price is
B(T)e
R(T)T
= 1 (4)
From (4), we can see that B(T) can be expressed as a function of R(T),
and conversely. Indeed, we can write
B(T) = e
R(T)T
(5)
and
R(T) =
log B(T)
T
(6)
Those relationships, (5) and (6),1 are shown in the right-hand side of g-
ure 12.1.
Consider now the forward rate. In section 11.4, we denoted the forward
rate decided upon at time t, for a loan starting at time T for an inn-
itesimal trading period, as f (t; T). Since today is time 0, we can denote it
as f (0; T) or more simply as f (T). The relationship between the discount
function B(T) and the forward rate is given by equation (28) in chapter
11, where t is now replaced by 0. With our simplied notation, it thus
reads as
B(T)e
r
T
0
f (u) du
= 1 (7)
from which B(T) can be immediately deduced:
B(T) = e
r
T
0
f (u) du
(8)
1 There is of course no surprise in seeing the minus sign in (6); the surprise would be not to
see one. Indeed, B(T) being smaller than 1, its natural logarithm (as the logarithm B(T) with
any base larger than 1 for that matter) is always a negative number. Since R(T) is positive, a
minus sign is warranted in front of log B(T)=T.
To determine the forward rate f (T) from B(T), rst take the logarithm
of (8):
log B(T) =
_
T
0
f (u) du
and dierentiate log B(T) with respect to T:
d
dt
log B(T) = f (T)
The forward rate is thus
f (T) =
d
dt
log B(T) (9)
The relationships between the discount function and the forward rate, (8)
and (9), are shown in the left-hand side of gure 12.1.
Finally, what about any relationship between the forward rate and the
spot rate? From (5) and (8), we can write
e
r
T
0
f (u) du
= e
R(T)T
(10)
Hence the spot rate R(T) is simply the average of the forward rates
between 0 and T:
R(T) =
_
T
0
f (u) du
T
(11)
From (10) or (11) we can write:
_
T
0
f (u) du = T R(T) (12)
Taking the derivative of (12) with respect to T yields:
f (T) = R(T) T
dR
dt
(13)
242 Chapter 12
The relationships between the forward rate and the spot rate, (11) and
(13), are shown at the bottom of gure 12.1. Thus there is indeed a one-
to-one relationship between each of the concepts of discount function,
spot rate, and forward rate. The moral: once you know one, you know the
other two, and there are always two ways of going from one to another
a direct way and an indirect one.
Two schools of thought have had a friendly battle these last 30 years
in the quest for the Holy Grail. We can rst distinguish the so-called
``parametric'' methods; their promoters' aim is to model, for instance, the
forward curve by a parametric function. The other contender are the
``spline'' methods (to be dened soon). We will now present those two
schools of thought, focusing on models that are relatively recent and that
are widely used by practitioners, especially by central banks. As the
reader might surmise, the most recent development in this eld, quite
interestingly, draws from both schools of thought.
Discount function =
price of zero-coupon bond
B(T)
Forward rate
f (T)
Spot rate =
zero-coupon yield
R(T)
f (T) = R(T) + T
dR
dT
f(T) = log B(T)
d
dT
R(T) =
(11)
f (u)du
T
0
T
B(T) = e
f (u)du
0
T
B(T) = e

R(T)T
(9)
R(T) =
log B(T)
T
(6)
(5) (8)
(12)
Figure 12.1
The one-to-one relationship between the discount function, the spot rate, and the forward
rate.
12.2.1 Parametric methods
We will rst examine the Nelson-Siegel (1987) model, and then turn to its
extension by Svensson (1994, 1995).
12.2.1.1 The Nelson-Siegel model
A very interesting model for the forward rate curve was proposed by
Charles Nelson and Andrew Siegel (1987).2 As we know, modeling the
forward rate curve is equivalent to modeling the spot rate curve (through
averaging the forward rate curve). The authors consider for the forward
rate quite an interesting function, which can give rise to practically all
shapes of spot curves observed on the markets. This is a Laguerre func-
tion3 plus a constant, the formula of which is
f (T) = b
0
b
1
e
T=t
1
b
2
t
1
Te
T=t
1
(14)
where T is the variable, and b
0
, b
1
, b
2
, and t
1
are parameters to be
estimated.
This expression implies the following equation for the spot rate. (See
equation (11) and also the bottom of gure 12.1.) Taking the average
value of the forward rates, we get
R(T) =
_
T
0
f (u) du
T
= b
0
(b
1
b
2
)
t
1
T
(1 e
T=t
1
) b
2
e
T=t
1
(15)
Let us evaluate limiting values of the spot rate and the forward rate
when T goes either to innity or to zero.
Consider rst the case where T y; from (15), R(T) tends toward the
constant b
0
. From (14), the same conclusion applies to the forward rate:
lim
Ty
f (T) = b
0
.4
Suppose now that the maturity T goes to zero. From what we know of
the spot rate as an average of forward rates, any limit when T 0 must
2 C. Nelson and A. Siegel, ``Parsimonious Modeling of Yield Curve,'' Journal of Business,
60, no. 4 (1987): 473489.
3 For a description of Laguerre functions, see for instance E. Courant and D. Hilbert,
Methods of Mathematical Physics (New York: Wiley, 1953), pp. 9397.
4 Notice that from lim f (T)
Ty
= b
0
, limR(T)
Ty
= b
0
, because R(T) is the average of
the f (u)'s. However, the converse would have been much harder to prove.
244 Chapter 12
be the same for either concept. Indeed, from (14) we have
lim
T0
f (T) = b
0
b
1
and the same is true for R(T); applying L'Ho pital's rule to the middle
term of the right-hand side of equation (15), we get
lim
T0
R(T) = b
0
b
1
We are now ready to examine the shapes that the average of a Laguerre
function can take, and we will make sure they t a great deal of observed
forms of the spot rate structure. We will try not to lose any generality in
the function (15) by setting b
0
= 4:5 (expressed in percent per year) and
t
1
= 1. Also, we will set b
1
= 1. Denoting by a the only parameter left,
we get from (15)
R(T) = 4:5 (1 a)
1 e
T
T
ae
T
(16)
We can check that limR(T)
Ty
= 4:5 (b
0
= 4:5), and that
limR(T)
T0
= 3:5(b
0
b
1
= 3:5).5 We have now given to parameter a
the values 2; 1; 0; 1; 3; 5. The resulting curves are depicted in gure
12.2. Indeed, we obtain the most familiar shapes of the spot rate structure.
One feature, however, is worth mentioning here because it is a little sur-
prising: the relatively slow convergence of the spot rates to their common
value (b
0
= 4:5) when T increases indenitely. Table 12.1 gives the values
we obtain with our values (in percent per year) for the Nelson-Siegel
model.
This slowness of convergence is even more apparent in the original
Nelson-Siegel model, where the authors considered a wider range of val-
ues for parameter a (from 6 to 12). (We have chosen to set their b
0
coecient to 3 in order to conform with their graph on page 476.) (Note
that table 12.2 does not carry the values for a = 1, for which maximum
convergence speed can be observed.)
5 There is probably a small typo on page 476 of the Nelson and Siegel article. To be consis-
tent with gure 1 of the paper, the rst coecient of the equation on the same page should be
3 (instead of 1). Or, if the equation is correct, then the rst node of the curves should be
at the origin, and all curves should be translated down by two.
0 5 10 15
(a) Original set of data points
20 25 30
0
6
5
4
3
2
1
Data
Svensson
Spline
0 5 10 15
(b) Change of single data point
20 25 30
0
6
5
4
3
2
1
Data
Svensson
Spline
Fig. 12.2
Comparison between the sensitivity of a spline regression and that of a parametric regression
to a change of one data point. (Reproduced from Nicola Anderson and John Sleath, ``New
Estimates of the UK Real and Nominal Yield Curves,'' Bank of England Quarterly Bulletin
(November 1999): 386, with the kind permission of the authors and the Bank of England,
Monetary Instruments and Markets Division.)
246 Chapter 12
12.2.1.2 The Svensson model
The Svensson model adds exibility to the Nelson-Siegel formulation by
adding a potential extra hump in the forward curve.6 Its basic equation
reads:
f (T) = b
0
b
1
e
T=t
1
b
2
T
t
1
e
T=t
1
b
3
T
t
1
e
T=t
2
(17)
which implies that six coecients ( b
0
, b
1
, b
2
, b
3
, t
1
, and t
2
) have to be
estimated. This model had considerable impact in the practitioners'
world: it was used between 1995 and the end of the twentieth century by
many central banks, most notably the Bank of England.
The highly innovative paper by Nicola Anderson and John Sleath7 (of
the Bank of England's Monetary Instruments and Markets Division)
explains very clearly why these authors looked for a new approach that
Table 12.1
Spot rate values for long maturities; b
0
F4:5; b
1
FC1
Maturity
(years) a = 2 a = 1 a = 0 a = 1 a = 3 a = 5
25 4.38 4.42 4.46 4.5 4.58 4.66
100 4.47 4.48 4.49 4.5 4.52 4.54
1000 4.497 4.498 4.499 4.5 4.502 4.504
Table 12.2
Spot rates for long maturities; b
0
FB3; b
1
FC1; original Nelson-Siegel values for a
Maturity
(years) a = 6 a = 3 a = 0 a = 3 a = 6 a = 12
25 2.72 2.84 2.96 3.08 3.2 3.32
100 2.93 2.96 2.99 3.02 3.05 3.08
1000 2.993 2.996 2.999 3.002 3.005 3.008
6 L. Svensson, `Èstimating and Interpreting Forward Interest Rates: Sweden 199294,'' IMF
Working Paper, No. 114 (1994); L. Svensson, `Èstimating Forward Interest Rates with the
Extended Nelson and Siegel Method,'' Sveriges Risbank Quarterly Review, 3 (1995): 13.
7 N. Anderson and J. Sleath, ``New Estimates of the UK Real and Nominal Yield Curves,''
Bank of England Quarterly Bulletin (November 1999): 384392.
would improve on the estimates of the Nelson-Siegel and Svensson
approaches, and why this approach is now in use at the Bank of England.
Apart from the appearance of additional information from the gilt
market,8 the main impetus for changing the in-house method of estimat-
ing the yield curve was the appearance of a new model developed by
Daniel Waggoner (1997), of the Federal Reserve Bank of Atlanta.9
Waggoner's work belongs to the family of ``spline'' methods that we will
now present.
12.2.2 Spline methods: the Fisher-Nychka-Zervos, the Waggoner, and the
Anderson-Sleath models
12.2.2.1 What is a spline?
From a mathematical standpoint, a spline is a piecewise polynomial,
which is a function made up of individual polynomial segments joined at
so-called ``knot points.'' At those knot points the function and its rst
derivative are continuous; note that the latter condition means the corre-
sponding curve is smooth in the sense that its curve does not make any
angle at any of its points. To the eye, a curve corresponding to a spline
will always look as if it represented one polynomial, albeit perhaps of a
high order.10
In fact, as stated in most nance texts that deal with the subject, most
splines are made up with piecewise cubic polynomials, and the reader
might ask why this is so and why at least some of the segments are not
second-order or fth-order polynomials, for instance. The answer to this
question is far from obvious and quite interesting. It certainly requires
going back to the origin of the word ``spline.'' We will then discover some
new and fascinating territory.
8 The gilt-edged market is the UK Government bond market. `Ìts origins,'' says the New
Palgrave Dictionary of Money and Finance (New York: Macmillan, 1992), p. 240 ``go back
to 1694 when the Bank of England was founded to help the Government raise money to ght
the French''(!) (Let us not forget, on the other hand, that the Banque de France was founded
by Napoleon to ght the rest of the world, and for a number of other no more commendable
purposes.)
9 D. Waggoner, ``Spline Methods for Extracting Interest Rate Curves from Coupon Bond
Prices,'' Working Paper, No. 97-10 (1997), Federal Reserve Bank of Atlanta.
10 Technical references to splines are from C. de Boor, A Practical Guide to Splines (New
York: Springer-Verlag, 1978), and G. Wahba, Spline Models for Observational Data (Phila-
delphia: SIAM, 1990).
248 Chapter 12
More than a century ago, the word ``spline'' had a number of dierent
meanings; among them is the following, which we extracted from the
Oxford English Dictionary (Vol. XVI, p. 283): `À exible strip of wood or
hard rubber used by draftsmen in laying out broad sweeping curves, espe-
cially in railroad work.'' When draftsmen designed in such a way curves
that had to pass through xed points, knowingly or not, they achieved
two results at the same time. First, they minimized the curvature of the
spline. Second, the lath took a form that minimized its deformation energy.
12.2.2.2 Some strangeand niceconsequences
Closing your favorite dictionary, it dawns on you that your younger sister
is majoring in physics (which of course makes you a little jealous, but you
know you can always count on her unless she is watching Beverly Hills).
So you ask her what the deformation energy is for a thin, homogeneous
lath whose function f (x) is dened over an interval [x
0
; x
n
[ going through
n 1 pivotal point (x
j
; y
j
), j = 1; . . . ; n. A little surprised by the sim-
plicity of your question, she answers that the deformation energy is, dis-
regarding a few constants, basically represented by the integral
I =
_
x
n
x
0
[ f
//
(x)[
2
dx (18)
Luck is still on your side. It turns out that you have been fortunate
enough to take a course on growth and investment.11 One of the instruc-
tor's pet hobbies was the calculus of variationsthe calculus by which
you look for extremals of functional relationships. These are relationships
between a function and a number equal to the integral of an expression
involving x, f (x) and its successive derivatives, for instance. Such a func-
tional may be written as
v[ y(x)[ =
_
b
a
F[x; f (x); f
/
(x); f
//
(x); . . . ; f
(n)
(x)[ dx (19)
So you recognize what your little sister has quickly scribbled on the
palm of your hand as nothing else than a particular case of (19), where all
11 Was it by any chance given by Olivier de La Grandville? That is a possibility.
arguments of F(:) are missing, except the fourth ( f
//
(x)). Furthermore,
F(:) in your case is ( f
//
(x))
2
. Now you remember that your instructor
highly praised an excellent, very readable text by L. Elsgolc12 that gives
the solution to the problem in the form of an Euler-Poisson equation: if
y(x) minimizes v[ y(x)[, it has to be a solution of the following Euler-
Poisson equation:
F
y

d
dx
F
y
/
d
2
dx
2
F
y
// (1)
n
d
n
dx
n
F
y(n)
= 0 (20)
where F
y
denotes the partial derivative of the integrand F with respect to
y, and F
y( j)
the partial derivative of F with respect to the derivative of
order j of y.
The Euler-Poisson equation is a dierential equation of order 2n. It
generalizes the solution discovered by Euler in 1744 to the following
problem: nd a curve y(x) that would be an extremum of
u[ y(x)[ =
_
b
a
F[x; f (x); f
/
(x)[ dx (21)
Euler found that if y(x) was such an extremal, it had to be the solution of
the second-order dierential equation
F
y

d
dx
F
y
/ = 0 (22)
Now why is the Euler-Poisson a dierential equation of order 2n? Since
Beverly Hills has not yet started, you turn again to your sister for some
sibling support. This time, deservedly, she gives you a bad rap. ``You
should be ashamed of yourself,'' she says. `Ì do remember seeing in your
class notes that your instructor (what was his funny name again?) had
12 L. Elsgolc, Calculus of Variations, International Series of Monographs in Pure and
Applied Mathematics (Reading, Mass.: Pergamon Press, 1962). Another excellent source is
by the same author (with a dierent spelling: L. Elsgolts), Dierential Equations and the
Calculus of Variations (Moscow: MIR Publishers, 1970).
250 Chapter 12
taken the trouble to write in full the Euler equation (22), which reads
F
y
(x; y; y
/
)
d
dx
F
y
/ (x; y; y
/
) = 0
or
qF
qy
(x; y; y
/
)
q
2
F
qxqy
/
(x; y; y
/
)
q
2
F
qyqy
/
(x; y; y
/
)y
/
q
2
F
qy
/2
(x; y; y
/
)y
//
_ _
= 0
(23)
``The last term of (23),'' your sister goes on, ``contains a function of
x, y, and y
/
multiplying the second derivative y
//
; hence the Euler equa-
tion is a nonlinear second-order dierential equation with nonconstant
coecients, generally depending not only on x but on y and y
/
as well. If
you do the same with the Euler-Poisson equation (20), you notice that you
have to take the nth-order derivative of an expression depending on the
nth-order derivative of yhence you are facing a 2n-order dierential
equation.
``What's more,'' she adds, ``your instructor gave you a way of obtaining
the Euler-Poisson equation through simple economic reasoning. To that
eect, he handed you an article of his. He could guarantee the quality of
the paper, he said, because you could use it to hold those greasy french
fries the next time you went to the ballpark. You thought it very funny to
do just that the last time you went to see the Giants take care of the
Dodgers, and now you pay the price.'' With that she slammed her door.13
12.2.2.3 Determining the order of the optimal polynomial spline
Ashamed of yourself as you may be, you are nevertheless ready to go
back to the functional minimizing the curvature as well as the deforma-
tion energy of the spline (18):
I[ y(x)[ =
_
x
n
x
0
[ f
//
(x)[
2
dx
Applying the Euler-Poisson equation (20), the spline should be a solution
13 We are grateful to your sister for having taken a few notes on this paper before you
headed for Pacic Bell Park. These can be found in appendix 12.A.2.
of the fourth-order dierential equation
d
2
dx
2
F
y
// = 0 (24)
In this case, F = (y
//
)
2
1[ f
//
(x)[
2
; therefore F
y
// = 2f
//
(x). The Euler-
Poisson is thus
d
2
dx
2
[2f
//
(x)[ = 2f
(4)
(x) = 0; (25)
f
(4)
(x) = 0 (26)
The most general solution of this simple dierential equation is obtained
by four successive integrations, which indeed lead to the third-order
polynomial
ax
3
bx
3
cx d = 0 (27)
A minimal convexity, minimal deformation energy curve will thus corre-
spond to a third-order polynomial.
To the best of our knowledge, J. H. McCulloch (1971 and 1975)14
was the rst author to propose using regression cubic splines to deter-
mine the discount function. He was followed by a number of authors who
have either improved the estimation techniques or set out to analyze po-
tential hereoskedasticity in the residuals. (The interested reader should
consult, in particular, Oldrich Vasicek and Giord Fong (1982);15 G.
Shea (1984);16 D. Chambers, W. Carleton, and D. Waldman (1992);17
and also T. S. Coleman, L. Fisher, and R. Ibbotson (1992).18 See also
14 J. H. McCulloch, ``Measuring the Term Structure of Interest Rates,'' Journal of Business,
44 (1971): 1931; J. H. McCulloch, ``The Tax-Adjusted Yield Curve,'' Journal of Finance, 30
(1975): 811830.
15 O. Vasicek and G. Fong, ``Term Structure Estimation Using Exponential Splines,''
Journal of Finance, 38 (1982): 339348.
16 G. Shea, ``Pitfalls in Smoothing Interest Rate Term Structure Data: Equilibrium Models
and Spline Approximations,'' Journal of Financial and Quantitative Studies, 19, no. 3 (1984):
253269.
17 D. Chambers, W. Carleton, and D. Waldman, `À New Approach to Estimation of Terms
Structure of Interest Rates,'' Journal of Financial and Quantitative Studies, 19, no. 3 (1984):
233252.
18 T. S. Coleman, L. Fisher, and R. Ibbotson, `Èstimating the Term Structure of Interest
from Data That Include the Prices of Coupon Bonds,'' Journal of Fixed Income (September
1992): 85116.
252 Chapter 12
the very innovative work of K. J. Adams and D. R. Van Deventer
(1994).19)
There was, however, one drawback to that approach. It implied undue
oscillations of the forward rate curve. In recent years, we have witnessed
three major successive breakthroughs to circumvent that problem. Each
of them displays remarkable ingenuity. These are the Fisher-Nychka-
Zervos (1995),20 the Waggoner (1997),21 and the Anderson-Sleath
(1999)22 models. We will describe them in order.
12.2.2.4 The Fisher-Nychka-Zervos model
The main objective of this powerful paper was twofold: rst, to price
bonds accurately and, second, to produce a relatively stable forward
curve. The originality of their contribution stems from the fact that the
spline is aimed directly at the forward rate curve. Furthermore, a
``roughness penalty'' is added in the process of minimizing the residual
sums of squares. (We will soon present this subject in more detail when
we describe the Waggoner model.) The results obtained by the authors for
daily (!) data from December 1987 through September 1994 are quite
impressive.
In an extremely detailed study, however, Robert Bliss (1997)23 observed
a slight problem with the Fisher-Nychka-Zervos method, in the sense that
it tended to misprice very short or short bonds. To try to correct that
slight defect, Daniel Waggoner (1997) suggested the following method.
12.2.2.5 An extension of the Fisher-Nychka-Zervos model: the
Waggoner model
Basically, Waggoner introduced a ``variable roughness penalty'' (VRP),
which we now describe.
19 K. J. Adams and D. R. Van Deventer, ``Fitting Yield Curves and Forward Rate Curves
with Maximum Smoothness,'' Journal of Fixed Income (June 1994): 5262.
20 M. Fisher, D. Nychka, and D. Zervos, ``Fitting the Term Structure of Interest Rates with
Smoothing Splines,'' Working Paper, No. 95-1 (1995), Finance and Economics Discussion
Series, Federal Reserve Board, 1995.
21 D. Waggoner, ``Spline Methods for Extracting Interest Rate Curves from Coupon Bond
Prices,'' Working Paper, No. 97-10 (1997). Federal Reserve Bank of Atlanta.
22 N. Anderson and J. Sleath, ``New Estimates of the UK Real and Nominal Yield Curves,''
Bank of England Quarterly Bulletin (November 1999): 384392.
23 R. Bliss, ``Testing Term Structure Estimation Methods,'' Advances in Futures and Options
Research, 9 (1997): 197231.
Pure arbitrage, as we presented it in previous chapters, is a simplied
description of what can happen in nancial markets, in the sense that a
number of imperfections, or biases, are introduced into the picture. Those
biases mainly take the form of taxes (we could even say tax systems) and
various transaction costs.
Following Waggonerwhere the author denotes the spot rate R(T) as
y(t) and the discount function B(T) as d(t)suppose that the cashows
of a given coupon-bearing bond are paid on times t
j
, j = 1; . . . ; K. The
theoretical value of a bond i, P
i
, is therefore given equivalently by any of
the following three expressions:
P
i
=
K
j=1
c
j
d(t
j
) =
K
j=1
c
j
e
t
j
y(t
j
)
=
K
j=1
c
j
e
_
t
j
0
f (u) du
(28)
and the pricing equation, where
^
P
i
denotes the observed value of the ith
bond24 and e
i
the error term, is
P
i
=
^
P
i
e
i
; i = 1; . . . ; n (29)
(supposing we consider a sample of n bonds).
As Waggoner explains clearly, the oscillation of a spline is often an
increasing function of the number of its nodes. This is disturbing, since
there are very few reasons for the forward curve not to be a horizontal
line for very long maturities.
These oscillations can be controlled by imposing a roughness penalty
(denoted l) in the regression process. The idea, originally introduced by
Fisher, Nychka, and Zervos (1995), is to choose an optimal spot rate
function R(T)denoted by Waggoner as c(t)that minimizes the sum
of the following two elements:
.
the sum of the square of error terms
e
2
i
=
n
i=1
P
i

^
P
i
[c(t)[
2
(30)
24 As many authors do, Waggoner uses as observed price the average between the bid and
ask quotes for the bond.
254 Chapter 12
.
the roughness penalty value l times the curvature (or, as we have seen, a
measure of the lath deformation energy) of the interest rate spline,25
l
_
t
K
0
[c
//
(t)[
2
dt.
The size of l acts as a trade-o between minimizing estimation errors
and minimizing curvature (or, equivalently, between maximizing accuracy
of t and minimizing smoothness). So we have to solve the following
problem:
min
c(t)
K
i=1
P
i

^
P
i
[c(t)[
2
l
_
t
K
0
[c
//
(t)[
2
dt
_ _
(31)
This is what Fisher, Nychka, and Zervos did using a generalized cross-
validation method. Using the Waggoner notation, this method consists of
minimizing the expression
g(l) =
RSS(l)
[n ye
p
(l)[
2
where
n 1number of bonds
RSS(l) 1sum of residual sum of squares (rst part of (31))
e
p
(l) 1eective number of parameters
y 1a ``cost'' or ``tuning'' parameter, set by Fisher, Nychka,
Zervos, and Waggoner to 2
The contribution of Waggoner was to introduce a variable roughness
penalty (a function depending on time, l(t)), so that one is now looking
for a function c(t) such that
min
c(t)
K
i=1
P
i

^
P
i
[c(t)[
2
l
_
t
K
0
l(t)[c
//
(t)[
2
dt
_ _
(32)
25 Independently, C. H. Reinsch (1971) and O. Vasicek, in the appendix of the K. Adams
and D. Van Deventer (1994) paper, showed, through variational methods, that directly
attacking the forward curve with a spline necessarily implied looking for a quartic (and not a
cubic) spline. (See C. H. Reinsch ``Smoothing by Spline Functions II,'' Numerische Mathe-
matik, 16 (1971): 451454.) We may then be facing a contradiction between what theory
dictates and what practitioners do.
This was a ne idea: introducing an ad hoc roughness penalty enabled
Waggoner to retain the much needed exibility at the short end, and to
damp oscillations for long maturities. The l(t) function retained by
Waggoner was the following:
l(t) =
0:1 0 Yt Y1
100 1 Yt Y10
100,000 t Z10
_
_
_
The rst question the reader could ask is: how sensitive is the optimal
function c(t) to the above ad hoc choice of l(t)? Quite surprisingly,
Waggoner found that the optimal c(t) was not very sensitive to quite
dierent choices of the l(t) function (the author chose step functions for
l(t) as well as continuous functions), with results that seem to improve
somewhat on those obtained in the Fisher-Nychka-Zervos model.
12.2.2.6 The Anderson-Sleath model
In their 1999 contribution, Nicola Anderson and John Sleath rst asked a
very interesting question: what is the main advantage of spline methods
over parametric methods? The answer is not so obvious. The principal
advantage of the former methods over the latter is that individual
segments of the spline can move almost independently of one another,
contrary to a parametric curve. Anderson and Sleath showed quite per-
suasively that whenever a shock aects the data at one end of the curve,
the spline is (quite naturally) modied at that end, while the curve
obtained through a parametric method can be severely aected, with no
good reason, at the other end. They give an excellent example to demon-
strate this, which they were kind enough to let us reproduce here.
In gure 12.2(a), a set of points (made up from a power function with
noise added) is regressed using both a cubic spline and a Svensson two-
hump functional form (equation (17)). The closeness between the two
curves is such that it is almost impossible to distinguish between them. We
now change the last point of the regression: while retaining the same
abscissa, its ordinate moves up from 5 to 5.5 (see the circled point in both
gures). Not only does the Svensson curve move practically on its whole
length, but it receives a hump at its short end, certainly a surprising and
unwarranted consequence of such an insignicant change in the data.
Anderson and Sleath modied the Waggoner model in important ways.
First, they weighed dierences between any bond and its theoretical value
256 Chapter 12
with the inverse of its modied duration (in order to penalize pricing
errors on bonds that are more sensitive to interest rate changes than
others). Second, they chose to estimate an optimal function of maturity
l(m) that would be such that
log l(m) = L (L S)e
m=m
where L; S, and m are parameters to be estimated. Finally, they made
prices direct functions of the forward rates.
So they chose to minimize the expression
X
s
=
N
i=1
P
i
p
i
(c)
D
i
_ _
2
_
M
0
l
t
(m)[ f
//
(m)[
2
dm (33)
where
D
i
1modied duration of bond i
c 1parameter vector of polynomial spline to be estimated
M1maturity of the longest bond:
26
Their results denitely seem to improve upon those obtained in the
earlier models. As the reader will notice, Anderson and Sleath draw upon
both schools of thoughtsplining and parameter estimation.
A nal word is in order. While the Fisher, Nychka, and Zervos method
postulates that l is constant accross maturities but variable from day to
day, the Waggoner and the Anderson and Sleath models allow l to vary
across maturities. One could think perhaps of a roughness penalty that
would share both properties.
Appendix 12.A.1 A proof of the central limit theorem and a determination of the
probability distribution of the dollar return on an asset
There are many ways to prove the central limit theorem. Here is one of
the most direct (inspired by S. Ross27 and adapted to the concepts and
26 We have taken the liberty of modifying very slightly their notation.
27 S. Ross, A First Course in Probability (New York: Macmillan, 1987), p. 346.
notation of this chapter). As in many cases, proving the theorem amounts
to showing that when n becomes large, the standardized variable of r
0; n
has a moment-generating function that converges toward e
l
2
=2
, that is,
the moment-generating function of the unit normal variable.
Let r
0; n
1r for short. We have
r =
n
t=1
r
t1; t
; with r
t1; t
= log(S
t
=S
t1
)
By hypothesis, the r
t1; t
's are independent and identically distributed;
moreover, E(r
t1; t
) = m and VAR(r
t1; t
) = s
2
< y; hence, from inde-
pendence of the r
t1; t
's, we have
E(r) = nm1m
VAR(r) = ns
2
1s
2
s(r) =
n
_
s 1s
The standardized variable of the sum is
r =
r nm
n
_
s
=
n
t=1
r
t1; t
nm
n
_
s
=
n
t=1
r
t1; t
m
n
_
s
_ _
=
n
t=1
Y
t
n
_
where Y
t
=
r
t1; t
m
s
is the standardized variable of r
t1; t
, with E(Y
t
) = 0
and VAR(Y
t
) = 1.
The moment-generating function of r is the nth power of the moment-
generating function of
Y
t
n
_
, itself j
Y
t
=
n
_
(l), equal to j
Y
t
l
n
_
_ _
.28 Thus we
have
j
r
(l) = j
Y
t
l
n
_
_ _ _ _
n
What we want to prove is that lim
ny
j
r
(l) = e
l
2
=2
or, equivalently, that
lim
ny
log j
r
(l) = l
2
=2. We have
log j
r
(l) = n log j
Y
t
l
n
_
_ _
=
log j
Y
t
l
n
_
_ _
n
1
1
L
l
n
_
_ _
n
1
28 In this simple proof, we suppose that the moment-generating function for Y
t
exists.
258 Chapter 12
where L(x) = log j
Y
t
(x). We have
L(0) = log j
Y
t
(0) = log 1 = 0
L
/
(0) =
j
/
Y
t
(0)
j
Y
t
(0)
= 0 (since j
/
(0) = 0)
L
//
(0) =
j
//
Y
t
(0)j
Y
t
(0) [j
/
Y
t
(0)[
2
j
//
Y
t
(0)
= 1 (since j
//
(0) = 1)
Let us apply L'Ho pital's rule to (34):
lim
ny
log j
r
(l) = lim
ny
L
/ l
n
_
_ _
ln
3=2
2n
2
= lim
ny
lL
/ l
n
_
_ _
2n
1=2
Applying L'Ho pital's rule to (35), we have
lim
ny
log j
r
(l) = lim
ny
l
2
2
L
//
l
n
_
_ _
=
l
2
2
This shows that r, the standardized variable of r, converges in distri-
bution toward the unit normal distribution N(0; 1).
Appendix 12.A.2 Deriving the Euler-Poisson equation from economic reasoning
In a highly insightful paper, Robert Dorfman (1969)29 showed how the
Pontryagin maximum principle (the fundamental theorem of optimal
control theory30) could be derived from pure economic reasoning, more
precisely by using capital theory. To that eect, he introduced the concept
of a modied Hamiltonian. Dierentiating this Hamiltonian with respect
to both the state variable and the control variable led him to the maxi-
mum principle. In this appendix, we will show how the Euler-Poisson
29 R. Dorfman, `Àn Economic Interpretation of Optimal Control Theory,'' The American
Economic Review (1969): 817831.
30 The original, technically demanding, work of Pontriaguine is in L. Pontryagin, V. Bol-
tyanskii, R. Gamkrelidze, and E. L. Mishchenko, The Mathematical Theory of Optimal
Processes (New York: Interscience Publishers, 1962). A deep (and dicult) discussion of
the relationship between the calculus of variations and optimal control theory is given in
Pontriaguine et al. (1962, chapter 5). A simpler presentation may be found in I. M. Gelfand
and S. V. Fomin, Calculus of Variations (Englewood Clis, N.J.: Prentice Hall, 1963) (see
Appendix II).
equation of the calculus of variations can be obtained by extending his
method to new Hamiltonians.
First, recall briey a simple version of Pontryagin's principle. Let k(t)
denote a state variable (for instance, the capital stock of a company or in
a country) and x(t) a control variable (that is, a variable that reects
decisions taken at time t). Suppose one wants to maximize (or minimize) a
functional such that
_
b
a
u(k
t
; x
t
; t) dt, where k
t
and x
t
stand for k(t) and
x(t) subject to the constraint
_
k = f (k
t
; x
t
; t)
Note that the control variable and the state variable have some eect
both on the integrand of the functional and on the rate of increase of the
state variable. Furthermore, these relationships may change in time.
Pontryagin's principle states that if one denes a function l(t) and a
Hamiltonian31
H = u(k
t
; x
t
; t) l(t) f (k
t
; x
t
; t)
then the optimal control time path x
+
(t) must be such that it solves the
following system of equations:
qH
qx
t
=
qu
qx
t
(k
t
; x
t
; t) l(t)
q f
qx
t
(k
t
; x
t
; t) (1)
qH
qk
t
=
qu
qk
t
(k
t
; x
t
; t) l(t)
qf
qk
t
(k
t
; x
t
; t) =
_
l(t) (2)
qH
ql
t
= f (k
t
; x
t
; t) =
_
k
t
(3)
For anyone familiar with dierential calculus and classical optimiza-
tion of functions only, there are at least three surprises in Pontryagin's
principle. The rst two come from the denition of the Hamiltonian,
which has little to do with a Lagrangian. First, the constraint
_
k =
f (k
t
; x
t
; t) does not appear in the form it has in a traditional Lagrangian.
Second, the familiar, constant Lagrange multiplier is here replaced by a
function of time to be determined. Finally, equation (2) may seem quite
astonishing
qH
qk
t
=
_
l(t)
_ _
.
31 There is no relationship between the l used in this appendix and the l used in the pre-
ceding chapter.
260 Chapter 12
To explain Pontryagin's principle from an economic point of view,
Dorfman had the stroke of genius to dene a new Hamiltonian that had
profound economic meaning, where the state variable and the control
variable would play symmetrical roles. Dorfman dened the total value of
the system at any point of time as
H
+
= u(k
t
; x
t
; t)
d
dt
[l(t)k(t)[ (4)
In this expression, he was adding to the integrand of the functional the
rate of increase of the state variable's value, itself equal to the product
of the amount of the state variable k(t) and its price l(t). H
+
could be
written as
H
+
= u(k
t
; x
t
; t)
_
l
t
k
t
l
t
_
k
t
(5)
It now makes a lot of sense to take the two partial derivatives of H
+
with
respect to k
t
and x
t
, respectively, and equate each of them to zero to
obtain
qH
+
qx
t
=
qu
qx
t
(k
t
; x
t
; t) l(t)
qf
qx
t
(k
t
; x
t
; t) = 0 (6)
qH
+
qk
t
=
qu
qk
t
(k
t
; x
t
; t)
_
l(t) l
t
qf
qk
t
(k
t
; x
t
; t) = 0 (7)
It is immediately apparent that equations (6) and (7) are equivalent to
Pontryagin's equations (1) and (2). Furthermore,
qH
+
ql
t
=
_
k
t
as in equation
(3).
Let us illustrate the power of Dorfman's (economic) Hamiltonian.
Suppose we want to address the classic, rst-order, variational problem
set by Jean Bernoulli in 169632 and solved in a general way by Euler half
32 The origin of the calculus of variations can be traced to 1696, when Jean Bernoulli
organized for his fellow mathematicians the following competition: let A and B be two xed
points in a vertical plane. Find the curve joining A and B such that a particle sliding along
the curve reaches B in minimum time. The solution is a cycloid (a curve generated by the
point of a circle rolling on a horizontal axis), and was found by Jean Bernoulli, his brother
Jacob, Newton, and the Marquis de L'Ho pital. Newton did not sign his solution (perhaps
fearing that H. M. Treasury, for which he was working at the time, would discover that he
was still dabbling in mathematics?). In any case, upon reading his solution, Jean Bernoulli
supposedly said, in awe: `Ì recognized the lion's paw.''
a century later (in 1744): nd k
t
(or
_
k
t
) such that the following functional
is minimized:
min
k
t
_
b
a
u(k
t
;
_
k
t
; t) dt
In this original variational problem, there are no constraints. One can
consider it a special case of an optimal control problem where the con-
straint
_
k
t
= f (k
t
; x
t
; t) is
_
k
t
= x
t
. Form then the modied Hamiltonian:
H
+
= u(k
t
;
_
k
t
; t)
_
l
t
k
t
l
t
_
k
t
(8)
Now take the derivatives of H
+
with respect to k
t
and
_
k
t
, and set both
equal to zero. We get
qH
+
qk
t
=
qu
qk
t
(k
t
; k
t
; t)
_
l
t
= 0 (9)
qH
+
q
_
k
t
=
qu
q
_
k
t
(k
t
;
_
k
t
; t) l
t
= 0 (10)
Dierentiate (9) with respect to time, and replace
_
l
t
in (8). The Euler
equation immediately appears:
qu
qk
t
(k
t
;
_
k
t
; t)
d
dt
qu
q
_
k
t
(k
t
;
_
k
t
; t) = 0 (11)
So this famous equation, which took so long to be found, can be derived
in three lines, thanks to Dorfman's modied Hamiltonian.
We should not stop at this point. It is only natural to ask whether this
method can be extended to obtain higher-order equations of the calculus
of variations. In particular, we should wonder whether we could obtain
the Euler-Poisson equation that we needed in this chapter to determine
the function corresponding to the optimal polynomial spline, that is,
minimizing both its curvature and the lath's deformation energy. The
problem amounted to nding y(t) such that the integral
_
b
a
[ y
//
(t)[
2
dt was
minimized.
So let us dene l
1
(t) as the price of y
t
and l
2
(t) as the price of y
/
t
and a
second-order Hamiltonian as
262 Chapter 12
H
(2)
= u(x; y; y
/
; y
//
)
d
dt
[l
1
(t)y
t
l
2
(t)y
/
t
[
= u(x; y; y
/
; y
//
) l
/
1
(t)y
t
l
1
(t)y
/
t
l
/
2
(t)y
/
t
l
2
(t)y
//
t
(12)
Take the partial derivative of H
(2)
with respect to y
t
; y
/
t
, and y
//
t
, and set
them equal to zero:
qH
(2)
qy
=
qu
qy
(x; y; y
/
; y
//
) l
/
1
(t) = 0 (13)
qH
(2)
qy
/
=
qu
qy
/
(x; y; y
/
; y
//
) l
1
(t) l
/
2
(t) = 0 (14)
qH
(2)
qy
//
=
qu
qy
//
(x; y; y
/
; y
//
) l
2
(t) = 0 (15)
Now take once the derivative of (14) with respect to time and twice the
derivative of (15). We get, respectively, with simplied notation,
d
dt
qu
qy
/
l
/
1
(t) l
//
2
(t) = 0 (16)
and
d
2
dt
2
qu
qy
//
l
//
2
(t) = 0 (17)
Substituting l
//
2
(t) from (17) into (16), we get
d
dt
qu
qy
/
l
/
1
(t)
d
2
dt
2
qu
qy
//
= 0 (18)
Finally, substituting l
/
1
(t) from (18) into (13), we obtain
qu
qy

d
dt
qu
qy
/

d
2
dt
2
qu
qy
//
= 0 (19)
which is nothing else than the Euler-Poisson equation, where the deriva-
tives of the function y = f (x) appear until the second order in the inte-
grand of the functional. Since minimizing the spline curvature or the
lath's deformation energy amounted to minimizing an integral containing
only the second derivative y
//
, the Euler-Poisson equation (19) boiled down
to
d
2
dt
2
qu
qy
//
= 0, which was equation (24) in our text.
This procedure can be generalized to any kind of variational problem.
Functionals involving all derivatives of y = f (x) up to the nth order can
be tackled in exactly this way. Also, this reasoning can be extended to
problems involving multiple integrals, where the integrand depends on
2n 1 arguments:
(a) n variables (x
1
; . . . ; x
n
),
(b) a function of these n variables z = f (x
1
; . . . ; x
n
), and
(c) the n partial derivatives
qf
qx
j
, j = 1, . . . , n.
The functional would thus be
_
. . .
_
R
F x
1
; . . . ; x
n
; z;
qz
qx
1
; . . . ;
qz
qx
n
_ _
dx
1
. . . dx
n
One would be led to the Ostrogradski equation, from which the Laplace
and Schro dinger equations, so central to physics, are derived. Finally, we
could apply this method to deal with functionals involving functions of n
variables and their partial derivatives up to the mth order. We would then
obtain generalizations both of the Euler-Poisson and the Ostrogradski
equations.33
Main formulas
.
Forward rate in the Nelson-Siegel model:
f (T) = b
0
b
1
e
T=t
1
b
2
t
1
Te
T=t
1
(14)
.
Spot rate in the Nelson-Siegel model:
R(T) =
_
T
0
f (u) du
T
= b
0
(b
1
b
2
)
t
1
T
(1 e
T=t
1
) b
2
e
T=t
1
(15)
.
Forward rate in the Svensson model:
R(T) = b
0
b
1
e
T=t
1
b
2
T
t
1
e
T=t
1
b
3
T
t
1
e
T=t
2
(17)
33 We take the liberty of referring the interested reader to our paper, ``New Hamiltonians for
Higher-Order Equations of the Calculus of Variations: A Generalization of the Dorfman
Approach,'' Archives des Sciences, 45, no. 1 (May 1992): 5158.
264 Chapter 12
.
Curvature or deformation of spline:
I =
_
x
n
x
0
[ f
//
(x)[
2
dx (18)
.
Euler-Poisson equation: to minimize
v[ y(x)[ =
_
b
a
F[x; f (x); f
/
(x); f
//
(x); . . . ; f
(n)
(x)[ dx (19)
a rst-order condition is that y(x) is a solution of
F
y

d
dx
F
y
/
d
2
dx
2
F
y
// (1)
n
d
n
dx
n
F
y(n)
= 0 (20)
which leads to a piecewise third-order polynomial minimizing I.
.
The Fisher-Nychka-Zervos model: choose optimal spot rate function
c(t) such that
min
c(t)
K
i=1
P
i

^
P
i
[c(t)[
2
l
_
t
K
0
[c
//
(t)[
2
dt
_ _
(31)
where P
i
are coupon-bearing prices and l is a roughness penalty value,
determined by generalized cross-validation method:
min
l
g(l) =
RSS(l)
[n ye
p
(l)[
2
where
n 1number of bonds
RSS(l) 1sum of residual sum of squares (rst part of (31))
e
p
(l) 1eective number of parameters
y 1a ``cost'' or ``tuning'' parameter, set by Fisher, Nychka,
Zervos, and Waggoner to two
.
The Waggoner model:
min
c(t)
K
i=1
P
i

^
P
i
[c(t)[
2
_
t
K
0
l(t)[c
//
(t)[
2
dt
_ _
(32)
.
The Anderson-Sleath model:
min
N
i=1
P
i
p
i
(c)
D
i
_ _
2
_
M
0
l
t
(m)[ f
//
(m)[
2
dm
where
D
i
1modied duration of bond i
c 1parameter vector of polynomial spline to be estimated
M1maturity of the longest bond
and
log l(m) = L (L S)e
m=m
Questions
12.1. Make sure that, in the Nelson-Siegel model, you can deduce the
spot rate structure (equation (15)) from the forward rate curve (equation
(14)).
12.2. You can nd the demonstration of the general Euler-Poisson equa-
tion in the texts on the calculus of variations, indicated in this chapter.
Make sure, however, that you can deduce, from an economic reasoning
point of view, the Euler-Poisson equation corresponding to
min
y(x)
_
x
1
x
0
F(x; y; y
/
; y
//
) dx
12.3. Make sure that you understand why
min
y(x)
_
x
1
x
0
[ f
//
(x)[
2
dx
leads to a third-order polynomial.
266 Chapter 12
13
ESTIMATING THE LONG-TERM EXPECTED
RATE OF RETURN, ITS VARIANCE, AND
PROBABILITY DISTRIBUTION
The key issue in investments is estimating expected return.
Fisher Black (1993)
In Memoriam (19411995)
Chapter outline
13.1 Notation 268
13.2 Estimating the expected value of the one-year return, its variance,
and its probability distribution 269
13.2.1 Determining E(log X
t1Y t
) and VAR(log X
t1Y t
) 269
13.2.2 What would be the consequences of errors in
parameters' estimates? 270
13.2.3 Recovering the expected value and variance of the yearly
rate of return, E(R
t1Y t
) and VAR(R
t1Y t
) 271
13.2.4 The probability distribution of the one-year return 272
13.2.4.1 Determination of the lognormal density
function and its relationship to the normal
density 276
13.2.4.2 Fundamental properties of the yearly rate of
return and their justication 277
13.3 The expected n-year horizon rate of return, its variance, and its
probability distribution 279
13.4 Direct formulas of the expected n-horizon return and its variance
in terms of the yearly return's mean value and variance 281
13.5 The probability distribution of the n-year horizon rate of return 283
13.6 Some concrete examples 284
13.6.1 Example 1 284
13.6.2 Example 2 285
13.7 The central limit theorem at work for you again 285
13.8 Beyond lognormality 288
Appendix 13.A.1 Estimating the mean and variance of the rate of
return from a sample 289
Appendix 13.A.2 Determining the expected long-term return when the
continuously compounded yearly return is uniform 291
Overview
We all wish we could know more about the future rate of return on any
kind of investment, be it an investment in stocks or one in bonds. Know-
ing more means having a precise idea about the expected value, its vari-
ance, and its probability distribution. The subject is important and comes
with a wealth of surprises. Suppose, for instance, that you set out to pro-
tect yourself against ination by buying a bond portfolio with a yearly
expected return of 5%, and the expected ination rate is also 5%; your
expected real return is zero. Suppose also that the annual real returns are
lognormally distributed, with a variance of 2.25%. It will probably sur-
prise you that your 30-year expected real return is minus 1.07% and that
the probability of having a negative 30-year real return is about 2/3!
In this chapter we will explain how we can determine the expected rate
of return on bond portfolios over a long horizon, the rate's variance, and
its probability distribution.1
13.1 Notation
We will use the following notation:
S
t
= an asset's value at end of year t
R
t1Y t
= the yearly rate of return, compounded once per year;
R
t1Y t
= (S
t
S
t1
)aS
t1
X
t1Y t
= S
t
aS
t1
= 1 R
t1Y t
= the yearly growth factor, or one plus
the yearly rate of return; in nancial circles it is called the
yearly ``dollar'' return
log X
t1Y t
= log(S
t
aS
t1
) = the yearly continuously compounded rate of
return
S
j
= the asset's value at the end of a trading day
R
j1Y j
= the daily rate of return, compounded once a day;
R
j1Y j
= (S
j
S
j1
)aS
j1
X
j1Y j
= S
j
aS
j1
= 1 R
j1Y j
= the daily growth factor; equivalently,
the daily ``dollar'' return, or one plus the daily return,
compounded once a day
1 Some of the material in this chapter has been drawn from our paper, ``The Long-
Term Expected Rate of Return: Setting It Right,'' Financial Analysts Journal (November/
December 1998): 7580, with kind permission of the Journal. Copyright 1998, Association
for Investment Management and Research, Charlottesville, Virginia. All rights reserved.
268 Chapter 13
log X
j1Y j
= log(S
j
aS
j1
) = the daily continuously compounded rate of
return
R
0Y n
= the n-year horizon rate of return; n, not necessarily an
integer, is expressed in years. R
0Y n
solves S
n
= S
0
(1 R
0Y n
)
n
;
R
0Y n
= (S
n
aS
0
)
1an
1.
X
0Y n
IS
n
aS
0
= the n-year horizon dollar return over a period of n
years
X
1an
0Y n
I(S
n
aS
0
)
1an
= the n-year horizon dollar return per year;
R
0Y n
= X
1an
0Y n
1
Y
(z) = E(e
zY
) = the moment-generating function of Y, where Y is
a continuous random variable
13.2 Estimating the expected value of the one-year return, its variance, and its
probability distribution
Our rst aim is to estimate the probability distribution of 1 R
t1Y t
=
X
t1Y t
and its rst two moments. A rst step may be to estimate the proba-
bility distribution of log X
t1Y t
and E(log X
t1Y t
), VAR(log X
t1Y t
). We will
be able to do this by having recourse to the central limit theorem. We will
then recover E(R
t1Y t
), VAR(R
t1Y t
) and its probability distribution. Of
course, we all know how little value or credence can be attributed to an
estimation of E(R
t1Y t
) based on past data. However, we should not for-
get that the procedure we describe is widely used for estimating variances
and covariances; we therefore keep it for the record and for reasons of
symmetry. Furthermore, the practitioner may well have his own method
of estimating the expected value of the continuously compounded return
and its variance; but there will still remain the question of recovering from
those estimates the expected value and variance of the yearly rate of return.
13.2.1 Determining E(log X
t1, t
) and VAR(log X
t1, t
)
Suppose we know the daily trading prices S
j
for a long period. We can
therefore determine over such a long period log(S
j
aS
j1
) the daily con-
tinuously compounded rates of return. For the time being, suppose we
have no indication as to the nature of the probability distribution of the
log(S
j
aS
j1
) variables. Assume, however, that these variables are inde-
pendent of one another and that they have nite variance. It is possible
269 Estimating the Long-Term Expected Rate of Return
to estimate both E[log(S
j
aS
j1
)[ Im and VAR[log(S
j
aS
j1
)[ = s
2
from
our sample, in the usual way.2 Consider now that there are v trading days
in a year (for instance, v = 250). We have
log(S
t
aS
t1
) =
v
j=1
log(S
j
aS
j1
) (1)
The central limit theorem ensures that log(S
t
aS
t1
) converges toward a
normal variable when v becomes large (250 is very large in this instance),
with mean vmI and variance vs
2
Io
2
. The important point to realize
here is that this applies whatever the probability distribution of
log(S
j
aS
j1
) may be. For the central limit theorem to apply, we need only
independence of the log(S
j
aS
j1
) and a variance for each that is nite.
We thus have
log(S
t
aS
t1
) = log(X
tY t1
) @ N(vmY vs
2
) IN( Y o
2
)
13.2.2 What would be the consequences of errors in parameters' estimates?
As the reader may surmise, errors in the estimation of the mean will be of
greater consequence than errors in estimating variances (or covariances in
the case of portfolios). As Vijay Chopra and William Ziemba put it very
well in their 1993 paper,3 the following question arises: ``How much
worse o is the investor if the distribution of returns is estimated with an
error? . . . Investors rely on limited data to estimate the parameters of the
distribution, and estimation errors are unavoidable'' (p. 55; our italics).
The authors answer this crucial question by showing that the ``percentage
cash equivalent loss'' for the investor (see their exact denition of this
concept) due to errors in means is about 11 times that for errors in vari-
2 In appendix 13.A.1 we recall why an unbiased estimate of a variance from a sample
necessitates adjusting the measured variance of the sample.
3 V. Chopra and W. Ziemba, ``The Eect of Errors in Means, Variances and Covariances on
Optimal Portfolio Choice,'' Journal of Portfolio Management (Winter 1993): 611; reprinted
in W. T. Ziemba and J. Mulvey, Worldwide Asset and Liability Modelling (Cambridge:
Cambridge University Press, Publications of the Newton Institute, 1998).
270 Chapter 13
ances and over 20 times that for errors in covariances. These results con-
rm those obtained earlier by J. Kallberg and W. Ziemba (1984), who
found that consequences of errors in means were approximately 10 times
more important than errors in variances and covariances, considered
together as a set of parameters.4 Also, the statistically minded reader will
nd interest in applications of mean estimation techniques in a series of
papers by R. R. Grauer and N. H. Hakansson.5
13.2.3 Recovering the expected value and variance of the yearly rate of return,
E(R
t1, t
) and VAR(R
t1, t
)
We now want to recover the implied expected value and variance of the
yearly rate of return R
t1Y t
= (S
t
S
t1
)aS
t1
= X
t1Y t
1.
Start with E(X
t1Y t
), equal to 1 E(R
t1Y t
). We can write
X
t1Y t
= e
log X
t1Y t
; log X
t1Y t
@ N( Y o
2
)
Denoting by
log X
t1Y t
(z) the moment-generating function of log X
t1Y t
at
point z, we can write
E(X
t1Y t
) = E(e
log X
t1Y t
) =
log X
t1Y t
(1) (2)
E(X
t1Y t
) can thus be seen to be the moment-generating function of
log X
t1Y t
at point 1.6 Since we have
4 J. G. Kallberg and W. T. Ziemba, ``Mis-specication in Portfolio Selection Problems,'' in
G. Bamberg and K. Spremann (eds.), Risk and Capital: Lecture Notes in Econometrics and
Mathematical Systems (New York: Springer-Verlag, 1989).
5 On these estimation techniques, see in particular, R. R. Grauer and N. H. Hakansson,
``Higher Return, Lower Risk: Historical Returns on Long-Run, Actively Managed Port-
folios of Stocks, Bonds and Bills, 19361978,'' Financial Analysts Journal, 38 (MarchApril
1982): 3953; ``Returns of Levered, Actively Managed Long-Run Portfolios of Stocks,
Bonds and Bills, 19341984,'' Financial Analysts Journal, 41 (SeptemberOctober 1985):
2443; ``Gains from International Diversication: 196885 Returns on Portfolios of Stocks
and Bonds,'' Journal of Finance, 42 (July 1987): 721739; and ``Stein and CAPM Estimators
of the Means in Asset Allocation,'' International Review of Financial Analysts, 4 (1995): 35
66.
6 There would have been two alternate ways of determining E(X
t1Y t
). The rst would have
been to take the integral
_
y
0
x f (x) dx, where f (x) is the lognormal density; the second one
would have been to calculate
_
y
y
exp(u)g(u) du, where u = log x and g(u) is the normal
density. The method we oer here, however, does not involve any calculations once we know
that the moment-generating function of a normal U = log X
t1Y t
@ N( Y o
2
) is
U
(z) =
E(e
zU
) = exp(z z
2
o
2
a2).
E(e
z log X
t1Y t
) =
log X
t1Y t
(z) = e
z
z
2
o
2
2
(3)
we get immediately
E(X
t1Y t
) =
log X
t1Y t
(1) = e

o
2
2
(4)
and
E(R
t1Y t
) = e

o
2
2
1 (5)
The variance of R
t1Y t
can be determined in an analogous way:
VAR(R
t1Y t
) = VAR(X
t1Y t
) = E(X
2
t1Y t
) [E(X
t1Y t
)[
2
(6)
We have
E(X
2
t1Y t
) = E(e
2 log X
t1Y t
) =
log X
t1Y t
(2) = e
2 2o
2
(7)
Therefore,
VAR(R
t1Y t
) = e
2 2o
2
e
2 o
2
= e
2 o
2
[e
o
2
1[ (8)
and
o(R
t1Y t
) = e

o
2
2
[e
o
2
1[
1a2
(9)
As an example, consider the case where estimates of and o
2
are
0.07905 and 0.03252, respectively. Applying (5), (8), and (9), we get
E(R
t1Y t
) = 10%, VAR(R
t1Y t
) = 4%, and o(R
t1Y t
) = 20%, respectively.
13.2.4 The probability distribution of the one-year return
Before rushing to the analysis of the n-year horizon rate of return, it will
be useful to consider carefully the probability distribution of the one-year
return. First, remember that, from the hypotheses we made on the daily
rate of return, the central limit theorem led us to conclude that the loga-
rithm of X
t1Y t
was normally distributed N(vmY vs) IN( Y o
2
). X
t1Y t
thus
272 Chapter 13
being lognormally distributed, suppose we wish to determine the proba-
bility that R
t1Y t
is between two values a and b. Keeping in mind that
log X
t1Y t
can be written as oZ,7 Z@ N(0Y 1), we can write
Prob(a ` R
t1Y t
` b) = Prob(1 a ` X
t1Y t
` 1 b)
= Prob[log(1 a) ` log X
t1Y t
` log(1 b)[
= Prob[log(1 a) ` oZ ` log(1 b)[
= Prob
log(1 a)
o
` Z `
log(1 b)
o
_ _
So, nally, we have
Prob(a ` R
t1Y t
` b) = p
log(1b)
o
_ _
p
log(1a)
o
_ _
(10)
where p(X) denotes the cumulative probability distribution of the unit
normal variable.
As a rst example of an application, let us consider our former case
where = 0X07905, o
2
= 0X03252 (o = 0X18033), implying E(R
t1Y t
) =
10%, and VAR(R
t1Y t
) = 4%, respectively. An apparently innocent ques-
tion is the following: what is the probability that the yearly rate of return
R
t1Y t
is below its expected value, equal to 10%? This is akin to deter-
mining the probability that R
t1Y t
is between a = 1 (100%, the mini-
mum it can reach even if the asset becomes worthless) and b = 0X1 (10%).
Applying (10), where a = 1 and b = 0X1, we get
Prob(1 ` R
0Y 1
` 0Y 1) = Prob[R
0Y 1
` E(R
0Y 1
)[
= p
log 1X1 0X07905

0X03252
_ _
p(y)
= p(0X0902) = 53X6%
So there is a slightly higher chance that the yearly return will be less than
its expected value (10%). This result, a direct consequence of the skewness
7 For notational convenience, we omit the subscript t 1, t and write Z
t1Y t
IZ.
of the lognormal distribution, is important; it will be of consequence when
we consider longer horizons, and therefore deserves explanation.
The fact that it is more likely that the yearly return will be lower than
its expected value (10%) has to be clearly understood. In our opinion, a
good starting point is to have a clear picture of the links between the
normal distribution and the lognormal distribution. (Remember that
the normal distribution governs U Ilog X
t1Y t
= log(S
t
aS
t1
) and that
the lognormal distribution is that of e
log X
t1Y t
= X
t1Y t
.) In gure 13.1 we
depict on the horizontal axis the probability distribution of U = log X
t1Y t
which is normal: U @ N( Y o
2
). In our case, = 0X07905 and o
2
=
3
2
N
o
r
m
a
l

d
e
n
s
i
t
y
Lognormal density
2
.
5 2
1
.
5 1
0
.
5
1
0.5
1.5
2
x
0
0 0.5 0.5
0 0.5 1 0.5 1
0
x = e
u
u = log x
2.5
e
+
2
/2
= mean = 1.1
e
= median = 1.0823 =
= 0.07905
1
geom.mean of X
Figure 13.1
A geometrical interpretation of three fundamental properties of the lognormal distribution:
(a) the expected value of X
tC1, t
Fe
U
Fe
msZ
(ZJN(0, 1)), E(X) Fe
ms
2
a2
is above the
median e
m
; (b) the probability that X
tC1, t
is below its expected value is above 50%; and (c) the
expected value of X
tC1, t
is an increasing function of the variance of UY s
2
. In this example,
E(log X) FE(U) F0X07905; s(log X) Fs
U
F0X18033.
274 Chapter 13
0X03252, or o = 0X18033. Now the exponential of U, e
U
or e
log X
t1Y t
=
X
t1Y t
is (by denition) lognormally distributed, and our purpose is to
discover the fundamental properties of the probability density of the log-
normal distribution of our variable of interest, X
t1Y t
, the yearly dollar
return on our investment.
Let us also draw in gure 13.1 the exponential x = e
u
, the possible
outcome values of the random variable X
t1Y t
= e
U
= e
log X
t1Y t
. Now
consider the average of the normal U @ N( Y o
2
), , which also is its
median. The exponential of , e
, will therefore be the median of the log-

normal, and we can already mark this value e
= e
0X07905
= 1X08226 on
the vertical axis x. On the other hand, we have determined through (4)
that the average of the lognormal was e
o
2
a2
, and therefore that the
average of the lognormal was higher than its median. For any continuous
random variable, its cumulative probability density to the left of its
median is 50%. Now if the expected value of X
t1Y t
is above the median, it
means that the probability for X
t1Y t
to be lower than its expected value is
50% plus the cumulative density between the median and the average.
This is precisely what is happening here: there is a 53.6% probability that
R(0Y 1) is lower than 10% (or that X
t1Y t
= 1 R(0Y 1) is lower than 1.1);
3.6% is exactly the cumulative density of the lognormal distribution
between the median of X
t1Y t
(e
= 1X0822) and the mean of X

t1Y t
(e
o
2
a2
= 1X1). Let us verify this by applying what we have just seen;
suppressing the lower indexes t 1, t for convenience, we have
Prob(median of X ` X ` mean of X)
= Prob(e
` X ` e
o
2
a2
) = Prob( ` log X ` o
2
a2)
= Prob( ` oZ ` o
2
a2) = Prob(0 ` oZ ` o
2
a2)
= Prob(0 ` Z ` oa2)
In our case, o = 0X18033; therefore,
Prob(median of X ` X ` mean of X)
= p(0X09017) p(0) = 0X536 0X5 = 0X036
= 3X6% as we wanted to ascertainX
Our purpose is now to explain and generalize these properties.
13.2.4.1 Determination of the lognormal density function and its
relationship to the normal density
It is easy to determine the lognormal density function once we know the
density of the normal function. Suppose that U = log X
t1Y t
is normally
distributed N( Y o
2
). In gure 13.1 we have represented this density in the
example chosen above, that is, with = 0X07905 and o = 0X18033 or o
2
=
0X03252. Let f (u) denote the normal density with average and standard
deviation o. We have
f (u) =
1
o

2
e
1
2
(
u
o
)
2
Now the lognormal distribution is the distribution of the exponential of
U @ N( Y o
2
), that is,8
X = e
U
From a realization of events point of view, each realized value u is related
to x by x = e
u
, or x(u) = e
u
.
To determine the density function of X, which we choose to denote as
g(x), we simply have to make sure that to each innitesimal probability
f (u) du corresponds an equal innitesimal probability g(x) dx. So we
must have
g(x) dx = f (u) du
which implies in turn9
g(x) = f (u)a
dx
du
(u)
where u is replaced, in the right-hand side, by u = log x and dxadu =

d(e
u
)adu = e
u
= x. Performing those substitutions, we have
g(x) = f (log x)ae
log x
= f (log x)ax
8 Notice how very unfortunate the standard terminology turns out to be: if U is normal, it
would have been logical to call e
U
the exponormal variable and not the lognormal variable
(i.e., the variable whose logarithm is normal, leaving the reader with the exercise of nding
that the lognormal is e
U
).
9 In order to ensure that the probability densities g(x) are always positive, we have to take
the absolute value
dx
du
. In our case, however, since x = e

u
, dxadu = x b 0 so we do not need
here the absolute value of the derivative.
276 Chapter 13
and therefore we get the lognormal density function
g(x) =
1
xo

2
e
1
2
[
log x
o
[
2
Both densities (the normal and the lognormal) are represented in gure
13.1, the normal on the horizontal axis u and the lognormal on the verti-
cal x. The relationship between x and u, x = e
u
has been drawn as well.
This diagram will be useful in understanding the important properties of
the lognormal.
13.2.4.2 Fundamental properties of the yearly rate of return and their
justication
Our purpose is now threefold. We want to show why
1. the expected value of the yearly dollar return, X
t1Y t
= e
U
= e
log X
t1Y t
is
higher than e
E(log X
t1Y t
)
= e
;
2. the probability of having a yearly dollar return smaller than its
expected value is higher than 50%; and
3. the expected value of the yearly dollar return is an increasing function
of the variance of the instantaneously compounded rate of return.
.
The expected value of the yearly dollar return is higher than its median
e
m
. We had observed earlier that (equation (4))
E(X
t1Y t
) = e
o
2
a2
This is clearly higher than e
, and it is a direct consequence of Jensen's

inequality, according to which, if h(U) is a convex transformation of U
(this is the case with X = h(U) = e
U
), then we always have
E(X) = E[h(U)[ b h[E(U)[
the proof of which can be found in most statistics books. But there is
an easy explanation of this inequality in the case where U has a symmet-
rical distribution (our case, since U = log X
t1Y t
is normal). Consider two
intervals of equal, arbitrary length I = o around on the horizontal u
axis (gure 13.1). To each interval I corresponds a given equal mass
probability under the normal unit curve; on the vertical x axis we can read
the value x = e
, as well as the intervals corresponding to the intervals I

dened previously on the abscissa. We notice immediately that the inter-
vals on the x axis are not equal, due to the convexity of the exponential
function x = e
u
. So the probability masses corresponding to the I inter-
vals, if represented by rectangles of equal areas (as they are on the left-
hand side of the diagram), will be of dierent heights, the height of the
upper rectangle being smaller than the height of the lower rectangle.
Where will the center of gravity of the mass of the two rectangles be
located on the x axis? Obviously to the right of the point corresponding to
the median, e
. Repeat this process for a new interval of the same length,

to the right and to the left of I and I. You will add, on the left-
hand side of the picture, two additional rectangles of equal area, the
upper one with a basis larger than the basis of the lower one. Therefore,
your center of gravity will be displaced to the right. Notice that each time
you repeat this process, you are adding intervals with mass probabilities
that decrease and tend toward zero. Henceforth, you may well surmise
that each time your center of gravity moves up on the x axis, it does so by
diminishing increments, tending toward a limit that turns out to be equal,
thanks to our previous calculations, to e
o
2
a2
.
.
The probability that the yearly dollar return is lower than its expected
value is higher than 50%. This is an immediate corollary of our last
property. Since the average of the lognormal X = e
U
, e
o
2
a2
is larger
than its median e
, it is clear that the probability of having an X value

smaller than E(X) is higher than 50%. In fact, the probability of X being
lower than E(X) is
Prob(X ` E(X) = e
o
2
a2
) = Prob(log X ` o
2
a2)
= Prob( oZ ` o
2
a2)
= Prob(Z ` oa2)
(A verication of our former calculations can be given here: with o
2
=
0X18033, prob(Z ` oa2 = 0X09017) = 53X6%, a result matching the one
we reached before.)
Note the simplicity of the result: the probability for the yearly return to
be lower than its expected value is simply the probability of the normal unit
variable to be lower than half of the standard deviation of the continuously
compounded return.
278 Chapter 13
This result has a nice geometrical interpretation and can be obtained
without recourse to any calculation. Refer again to gure 13.1, and watch
out for the values of the median and the mean of the lognormal, equal to
e
and e
o
2
a2
, respectively. The probability that X will be below E(X) is
the probability that log X will be below E(log X) = o
2
a2. This prob-
ability can be recovered by going back to the normal curve through the
log transformation on the diagram (the reciprocal of the exponential).
This is just the area spanned by the normal curve below o
2
a2, or the
area below Z = ( o
2
a2 )ao = oa2.
We will now explain why this probability is an increasing function of
the standard deviation of the continuously compounded rate of return.
.
The expected value of the yearly dollar return is an increasing function of
the variance of the continuously compounded return. From equation (4),
we notice a positive dependence between the variance o
2
of the con-
tinuously compounded rate of return and the expected yearly returnand
a powerful dependence at that. Our purpose is now to explain why this
is so.
Let us come back once more to gure 13.1. We can see immediately
that if U = log X
t1Y t
has a larger variance than before, and if we consider
the same mass probability as the one that corresponded to the I intervals,
our I intervals this time will have to be broader, entailing a larger dis-
crepancy in the bases of the two rectangles in the left part of the diagram.
Hence the center of gravity will be farther away from e
than it was when

the variance of U was lower. The same argument carries to additional
intervals; hence the center of gravity (the expected value of X
t1Y t
) is def-
initely increasing with the variance of U.
13.3 The expected n-year horizon rate of return, its variance, and its probability
distribution
Suppose that n is the length of our horizon, expressed in years; n is not
necessarily an integer number. For instance, if one year has 250 trading
days, as in our example before, n = 2X2 years means two years plus
(0X2)(250) = 50 trading days. Our n-year horizon yearly rate of return,
denoted R
0Y n
, is such that
S
0
(1 R
0Y n
)
n
= S
n
(11)
and is equal to
R
0Y n
=
S
n
S
0
_ _
1an
1 IX
1an
0Y n
1 (12)
where X
0Y n
= S
n
aS
0
.
We have
log X
0Y n
= log S
n
log S
0
=
nX250
j=1
log(S
j
aS
j1
) @ N(nX250m; nX250s
2
)
IN(n ; no
2
) = n

n
oZY Z@ N(0Y 1) (13)

Indeed, from our hypotheses the random variable log X
0Y n
turns
out to be normally distributed with mean nX250mIn and variance
nX250s
2
Ino
2
.
We can then write
E(X
1an
0Y n
) = E[(e
log X
0Y n
)
1an
[ = E(e
1
n
log X
0Y n
) (14)
The term on the far right of (14) is simply the moment-generating
function of log X
0Y n
at point 1an. We therefore have
E(X
1an
0Y n
) =
log X
0Y n
1
n
_ _
= e
1
n
n
1
2
1
n
2
no
2
= e

o
2
2n
(15)
and
E(R
0Y n
) = e

o
2
2n
1 (16)
Consider, as before, our estimates of = 7X905% per year and o
2
=
3X252%ayear
2
, that is, o = 18X003% per year. We have seen that they
entailed a yearly expected return E(R
0Y 1
) equal to 10%, and a standard
deviation of R
0Y 1
equal to 20%. Should our horizon be 2.2 years, we
would have
E(R
0Y 2X2
) = e
0X07905
0X03252
4X4
1 = 9X03% per year
Let us generalize this method to the determination of the kth moment
of X
0Y n
; we will use this result to determine the variance of R
0Y n
in a
straightforward way.
280 Chapter 13
The kth moment of X
1an
0Y n
can be written as
E(X
kan
0Y n
) = E(e
k
n
log X
0Y n
) =
log X
0Y n
(kan) (17)
and is thus equal to the moment-generating function of log X
0Y n
at point
kan; it is therefore equal to
E(X
kan
0Y n
) = e
k
n
n
1
2
k
2
n
2
no
2
= e
k
k
2
2n
o
2
(18)
As an application, let us determine the variance of the n-year horizon
expected rate of return.
VAR(R
0Y n
) = VAR(X
1an
0Y n
1) = VAR(X
1an
0Y n
) = E(X
2an
0Y n
)[E(X
1an
0Y n
)[
2
(19)
Using (18), we have
E(X
2an
0Y n
) = e
2
2o
2
n
(20)
As a consequence
VAR(R
0Y n
) = e
2
2o
2
n
e
2
o
2
n
(21)
and therefore
VAR(R
0Y n
) = e
2
o
2
n
(e
o
2
n
1)
The standard deviation of R
0Y n
is
o(R
0Y n
) = e

o
2
2n
(e
o
2
n
1)
1a2
(22)
Pursuing our previous example, we have
o(R
0Y 2X2
) = e
0X07905
0X03252
4X4
(e
0X07905
1) = 8X96%
13.4 Direct formulas of the expected n-horizon return and its variance in terms of
the yearly return's mean value and variance
It would, of course, be very convenient to express both the expected value
and the variance of the n-horizon return directly as functions of the one-
year expected return and variance. Let E(X
0Y 1
) = E(R
0Y 1
1) IE and
VAR(X
0Y 1
) = VAR(R
0Y 1
1) = VAR(R
0Y 1
) IV. We would like to express
formulas (16) and (21) directly in terms of E and V, which pertain to the
yearly return. Recall that we have
E IE(X
0Y 1
) = e

o
2
2
(4)
and
V IV(X
0Y 1
) = e
2 o
2
(e
o
2
1) = E
2
(e
o
2
1) (8)
From system (4) and (8) we can easily determine and o
2
:
= log E
1
2
log 1
V
E
2
_ _
(23)
o
2
= log 1
V
E
2
_ _
(24)
Plugging (23) and (24) into (16), (21), and (22), we get
E(R
0Y n
) = E (1 VaE
2
)
1
2
(
1
n
1)
1 (25)
VAR(R
0Y n
) = E
2
(1 VaE
2
)
(
1
n
1)
[(1 VaE
2
)
1
n
1[ (26)
which also leads to
o(R
0Y n
) = E (1 VaE
2
)
1
2
(
1
n
1)
[(1 VaE
2
)
1
n
1[
1a2
(27)
Using (25), (27) can be expressed simply as
o(R
0Y n
) = [1 E(R
0Y n
)[[(1 VaE
2
)
1
n
1[
1a2
(28)
As a check, note that with n = 1, E(R
0Y 1
) = E 1, VAR(R
0Y 1
) =
VAR(X
0Y 1
) = V, and that with n y, VAR(R
0Y y
) = 0, as they should.
282 Chapter 13
In the limit, the innite horizon expected return is
E(R
0Y y
) = e
1 = E 1
V
E
2
_ _
1a2
1 =
E
2
(E
2
V)
1a2
1 (29)
For instance, with E = 1X1 and V = 0X04, the innite horizon expected
return is E(R
0Y y
) = 8X23%. It is well known that the sample geometric
mean of a series of independent random variables converges asymptoti-
cally, by decreasing values, toward the geometric mean of the random
variable (see Hines10 ). Here E(X
1an
0Y y
) = E(R
0Y y
) 1 = E
2
(E
2
V)
1a2
is precisely the geometric mean of the probability distribution of the
yearly dollar return X
t1Y t
. So the innite horizon expected rate of return
(29) is simply the geometric mean of the lognormal minus one.
13.5 The probability distribution of the n-year horizon rate of return
Coming back to one of our basic formulas, (13), we can see that one plus
the n-horizon rate of return is a lognormal distribution with parameters
and o
2
an; that is, the logarithm of the n-year horizon dollar return,
log(1 R
0Y n
), is normally distributed N( Y o
2
an). This implies that
log(1R
0Y n
)
oa

n
is N(0Y 1); therefore, the probability that R
0Y n
is between any
two values a and b is p
log(1b)
oa

n
_ _
p
log(1a)
oa

n
_ _
, where p is the
cumulative probability distribution of the unit normal variable.
More formally, keeping in mind that X
1an
0Y n
= 1 R
0Y n
and that, from
(13), log X
1an
0Y n
= log(1 R
0Y n
) can be written as
o

n
Z, Z@ N(0Y 1), we
have
Prob(a ` R
0Y n
` b) = Prob(1 a ` 1 R
0Y n
` 1 b)
= Prob[log(1 a) ` log(1 R
0Y n
) ` log(1 b)[
= Prob[log(1 a) `
o

n
Z ` log(1 b)[
= Prob
log(1 a)
oa

n
` Z `
log(1 b)
oa

n
_ _
10 W. G. S. Hines, ``Geometric Mean,'' in S. Kotz and N. L. Johnson (eds.), Encyclopedia of
Statistical Science, vol. 3 (New York: Wiley, 1983), pp. 397450.
Therefore, we have
Prob(a ` R
0Y n
` b) = p
log(1 b)
oa

n
_ _
p
log(1 a)
oa

n
_ _
(30)
Conversely, we can easily determine an interval for R
0Y n
spanning a
given probability interval, for instance 95%. We can write
Prob( 1X96oa

n
` log(1 R
0Y n
) ` 1X96oa

n
) = 0X95 (31)
or, equivalently,
Prob(e
1X96oa

n
1 ` R
0Y n
` e
1X96oa

n
1) = 0X95 (32)
Thus, a 95% probability interval for the n-horizon rate of return is given
by (e
1X96o

n
1Y e
1X96o

n
1); in turn, and if needed, this interval can

be expressed directly in terms of E and V, using (23) and (24). We will
now give examples of applications.
13.6 Some concrete examples
13.6.1 Example 1
As a rst example, let us verify the exactness of the gures we have given
in our introduction. By hypothesis, the one-year expected real return
on the portfolio is 0%, and its variance is 2.25%. So we can plug
E = 1 0% = 1, V = 0X0225, and n = 30 into formula (25), to yield
E(R
0Y 30
) = 1X07%. The probability that R
0Y n
is negative is given by
Prob(R
0Y n
` 0) = Prob(1 R
0Y n
` 1)
Since log(1 R
0Y n
) is normal N( Y o
2
an), we rst have to determine
and o
2
from (23) and (24). We get
= log E
1
2
log 1
V
E
2
_ _
= 0X011125
and
o
2
= log 1
V
E
2
_ _
= 0X0222506; o = 0X149166
284 Chapter 13
Now we have
Prob(1 R
0Y n
` 1) = Prob[ log(1 R
0Y n
) ` 0[
= Prob
log(1 R
0Y n
)
oa

n
`

oa

n
_ _
= Prob Z `

oa

n
= 0X4085
_ _
Y Z@ N(0Y 1)
= p(0X4085) = 0X658
as stated in the introduction to this chapter.
13.6.2 Example 2
Let us now consider the case E(R
0Y 1
) = 10%; o(R
0Y 1
) = 20%. We then
have E IE(X
t1Y t
) = 1X1. Using (25), E(R
0Y 10
) turns out to be 8.4%;
similarly the standard deviation of R
0Y 10
can be calculated, using (28), as
6.2%. Suppose now that we want to know the probability for the 10-year
horizon to be negative. Referring to our results in section 13.4, we rst
have to translate E(= 1X1) and V(= 0X04) in terms of and o. Using (23)
and (24) respectively, we get = 7X90%, o
2
= 3X25%, o = 18X03%. The
probability for R
0Y 10
to be lower than b = 0 is equal to p
log(1b)
o

n
_ _
b=0
=
p
7X90%
18X03%a

10
_ _
= p(1X38) e8%. Similarly, the 95% prediction interval
for R
0Y 10
is (e
1X96oa

n
1Y e
1X96oa

n
1) = (3X2%Y 21X0%). On the

other hand, R
0Y 10
has a 68% probability of being between 2.2% and
14.6%.
13.7 The central limit theorem at work for you again
Suppose you are not quite sure about the exact nature of the probability
distribution of the yearly continuously compounded rate of return,
log X
t1Y t
. In particular, you are not certain it is normal, which implies
that you are not sure about the lognormality of X
t1Y t
. Nevertheless, you
go ahead and perform the calculations of the long-term expected return
E(R
0Y n
) as indicated before, assuming that the X
t1Y t
's are lognormal.
Suppose you want to know the type of error you have been making by
relying on a distribution (the lognormal) that is not necessarily the true
one. In order to have an idea of the error made, let us conduct the
following experiment. Assume that the true probability distribution of
log X
t1Y t
was uniform. As shown in appendix 13.A.2, this implies that the
underlying probability density of X
t1Y t
is hyperbolic, indeed quite a de-
parture from the lognormal, and that the expected value of 1 R
0Y n
(denoted Y
0Y n
) is
E(Y
0Y n
)[
log X
t1Y t
@ unif orm[aY b[
=
n
2

3
o
(e
(

3
o)an
e
(

3
o)an
)
_ _
n
=
n
b a
(e
ban
e
aan
)
_ _
n
(33)
where a =

3
o and b =

3
o.
This formula looks exceedingly dierent from the result corresponding
to the normal distribution for log X
t1Y t
:
E(Y
0Y n
)[
log X
t1Y t
@ N( Y o
2
)
= e

o
2
2n
= e
ba
2

(ba)
2
24n
(34)
which applies in the case of the normal distribution. Now try it in the
following example: let E(log X
t1Y t
) = = 10% and o(log X
t1Y t
) = o =
20%, and apply (14) and (15). The surprise comes when it turns out that
the rst four digits of the long-term expected return E(R
0Y n
) = E(Y
0Y n
) 1
are the same in each case when the horizon is as low as two years! (See
table 13.1.) Six digits are caught with ve years. The explanation of this
remarkable result lies of course in the strength of the central limit theo-
rem. Indeed, Y
0Y n
, the geometric average of the dollar yearly returns, can
Table 13.1
Comparison of E(R
0Y n
) when log X
tC1, t
is either normal or uniform, with E(log X
tC1, t
) F10%
and s(log X
tC1, t
) F20%
Horizon n
(years)
E(R
0Y n
) with
log X
t1Y t
@ normal
E(R
0Y n
) with
log X
t1Y t
@ uniform
2 0.116278 0.116267
4 0.110711 0.110709
6 0.108861 0.108861
8 0.107937 0.107937
10 0.107383 0.107383
20 0.106277 0.106277
30 0.105908 0.105908
100 0.105392 0.105392
286 Chapter 13
be written as
Y
0Y n
=
n
t=1
X
1an
t1Y t
_ _
= e
1
n
n
t=1
log X
t1Y t
(35)
and the central limit theorem can be applied to the power of e. We thus
have
Y
0Y n
ee
E(log X
t1Y t
)
o(log X
t1Y t
)

n
Z
Y Z@ N(0Y 1) (36)
and the expected value of Y
0Y n
is11
E(Y
0Y n
) ee
E(log X
t1Y t
)
o
2
(log X
t1Y t
)
2n
(37)
It turns out that in most cases (and ours is such a case) the average in
the power of e converges extremely quickly. (For a very good illustration
of the power of the central limit theorem, see the spectacular diagrams in
J. Pitman (1993).12) Hence, the result we have here: very early the distri-
bution of Y
0Y n
becomes lognormal, even if the continuously compounded
return log X
t1Y t
is far from normal (a uniform distribution in our case).
We should add that the stability of the mean reects the small annual
variance, as well as the central limit theorem.
In section 13.5 we promised to determine the limit of the long-term
dollar return when the horizon tends to innity; this is now easy to do,
thanks again to the central limit theorem. Indeed, turning to (36), it is
immediately clear that when n y, the limit is e
E(log X
t1Y t
)
, which is none
other then the geometric mean of the distribution of X
t1Y t
. (On this, see
Hines (1983).) In our example, the distribution of X
t1Y t
is hyperbolic if
log X
t1Y t
is uniform, lognormal if log X
t1Y t
is normal. It turns out that
this property can be extended to all moments of Y
0Y n
: each kth moment
of Y
0Y n
tends toward the kth geometric moment of the X
t1Y t
distribution.
(For a denition and properties of geometric moments of order k, see
O. de La Grandville (2000).13)
We can conclude that what matters most when forecasting expected
long-term returns are the rst two moments of the yearly returns, not their
11 To show this, use again the fact that E(Y
0Y n
) is a linear transformation of the moment-
generating function of Z, where
o(log X
t1Y t
)
n
plays the role of the parameter.
12 J. Pitman, Probability (New York: Springer-Verlag, 1993), pp. 199202.
13 O. de La Grandville, `Ìntroducing the Geometric Moments of Continuous Random
Variables, with Applications to Long-Term Growth Rates,'' Research Report, University of
Geneva, Geneva, Switzerland, 2000.
probability distribution. Whenever you assume yearly returns to be inde-
pendent and to have nite variance, do not worry about their probability
densities, and let the central limit theorem do the work for you.
13.8 Beyond lognormality
The lognormality hypothesis is indeed very convenient mathematically.
We must stress, however, that the lognormality assumption for asset prices
rests on the central limit theorem, which itself requires two sucient (but
not necessary) conditions14: rst, independence of the continuously com-
pounded returns; second, identical distributions for those returns. On the
other hand, all market participants are well aware that the market sys-
tematically exaggerates up and down movements when it experiences
huge swings; therefore, this independence condition is not always met.
Researchers have been conscious of this fact and have tried to model
the market's behavior by using so-called ``fat-tailed distributions'' (for
instance, Pareto-Levy distributions). Those distributions have in common
the fact that they do not possess nite moments (the corresponding inte-
grals do not converge), and they are known by their characteristic func-
tions only. Therefore, it becomes much more dicult to work with them
than with the convenient lognormal distribution. The rst person to con-
sider these possibilities was the mathematician Beno t Mandelbrot, in his
paper ``The Variation of Certain Speculative Prices''.15 His further work,
as well as other important contributions in this area, are well referenced
in his recent book Fractals and Scaling in Finance: Discontinuity, Concen-
tration, Risk.16
No doubt advances will still be made in this subject. Although we have
not applied these distributions in this chapter, we think that if it had been
the case, the long-term expected return would still have been smaller than
the one-year expected return, with a dierence of a magnitude compara-
ble to that obtained with the lognormal distribution. An interesting ques-
14 Note that the central limit theorem is valid under much wider conditions: changing dis-
tributions are allowed under certain conditions and under some degree of dependence be-
tween the log-returns, but not too much so.
15 B. Mandelbrot, ``The Variation of Certain Speculative Prices,'' Journal of Business, 36
(1993): 394419.
16 B. Mandelbrot, Fractals and Scaling in Finance: Discontinuity, Concentration, Risk (New
York: Springer, 1997).
288 Chapter 13
tion, of course, would be to know how much the skewness introduced in
these distributions inuences this dierence.
Appendix 13.A.1 Estimating the mean and variance of the rate of return from a
sample
Let X designate a random variable with unknown mean and unknown
variance o
2
. All possible values of X are supposed to be independent from
each other. Let (x
1
Y F F F Y x
n
s
) be a sample of size n
s
of this random vari-
able. For instance, X may be the continuously compounded daily return
on an asset (X = ln(S
j
aS
j1
) when S
j
is the asset's value at end of day j; x
j
may be its sample value measured at end of day j: x
j
= ln(S
j
aS
j1
)). We
now consider two statistics corresponding to this sample: its mean, x, and
its variance, s
2
. Its mean, x =
1
n
s
n
s
i=1
x
i
Y can be considered as one sample
value of a random variable that is the average of n
S
random variables X
i
:
X =
1
n
s
n
s
i=1
X
i
The expected value of X is E(X) =
1
n
s
n
s
i=1
E(X
i
) =
1
n
s
n
s
E(X) = E(X) =
. This means that we can consider the sample's average as representative
of the population's average. Let us now consider the variance of the
random variable X:
VAR(X) =
1
n
2
s
n
i=1
VAR(X
i
)
(since the X
i
variables are independent)
=
1
n
2
s
n
s
VAR(X
i
) =
VAR(X)
n
s
=
o
2
n
s
(since all variances of the X
i
's are common and equal to the (unknown)
VAR(X) = o
2
).
So the statistic X has an expected value that is the mean of the popu-
lation; its variance is o
2
an
s
, which converges to zero when n
s
y. The
larger the sample, the more condent we can be that the sample's average
is close to the true population's mean.
Consider now another statistic: the variance of the sample,
s
2
=
1
n
s
n
s
i=1
(x
i
x)
2
This is just one outcome value of a random variable that may be written
as
S
2
=
1
n
s
n
s
i=1
(X
i
X)
2
Let us now consider the expected value of S
2
, E(S
2
):
E(S
2
) =
1
n
s
E
n
s
i=1
(X
2
i
2X
i
XX
2
)
_ _
=
1
n
s
E
n
s
i=1
(X
2
i
)n
s
X
2
_ _
=
1
n
s
n
s
i=1
E(X
2
i
)n
s
E(X
2
)
_ _
= E(X
2
i
)E(X
2
) = o
2
2
E(X
2
)
We can write
E(X
2
) =
1
n
2
s
E(X
i
X
n
s
n
s
)
2
=
1
n
2
s
E[X
i
X
n
s
[
2
E[2n
s
(X
i
X
n
s
)[ E(n
2
s
2
)
In the above expression, the rst expectation is the expected value of a
quadratic form, which is equal rst to the n variances of X
i
, F F F , X
n
s
(hence to n
s
o
2
), to which one has to add n
s
(n
s
1) covariances, all equal
to zero, since the X
i
's are independent. The second expected value is zero,
since E(X
i
) = 0, i = 1, F F F , n
s
; the third one is just n
2
s
2
.
We have
E(X
2
) =
1
n
2
s
(n
s
o
2
n
2
s
2
) =
o
2
n
s

2
(Note that this last expression does make sense: since X approaches
when n y, it is reasonable to think that the expectation of X
2
should
approach
2
when n y. From the above expression, we can see that
this indeed is the case.)
290 Chapter 13
Finally, for E(S
2
) we get
E(S
2
) = o
2

2
o
2
n
s

2
=
n
s
1
n
s
o
2
In order to get an estimate of the true variance, we will therefore con-
sider the statistic
n
s
n
s
1
S
2
, denoted S
+
, whose expected value will be o
2
:
E(S
+
) = E
n
s
n
s
1
S
2
_ _
=
n
s
n
s
1
n
s
1
n
s
o
2
= o
2
Appendix 13.A.2 Determining the expected long-term return when the continuously
compounded yearly return is uniform
If log X
t1Y t
is uniform on [aY b[, we have the following relationships:
E(log X
t1Y t
) I =
a b
2
(1)
o(log X
t1Y t
) Io =
b a
2

3
(2)
Conversely, a and b are related to and o by
a =

3
o (3)
b =

3
o (4)
Denote for simplicity U = log X
t1Y t
; if h(u) is the density function of
U and f (x) is the density function of X
t1Y t
, we have the well-known
relationship f (x) = h(u)
du
dx
=
1
ba
1
x
. Hence the density function of X
t1Y t
is a hyperbola dened between e
a
and e
b
.
E(Y
0Y n
) is equal to
E(Y
0Y n
) = E
n
t=1
X
1an
t1Y t
_ _
=
n
t=1
E(X
1an
t1Y t
) = [E(X
1an
t1Y t
)[
n
(from the independence of the X
t1Y t
's). Evaluate now each E(X
1an
t1Y t
) in
this case:
E(X
1an
t1Y t
) = E(e
1
n
log X
t1Y t
) = E(e
1
n
U
)
=
_
b
a
e
1
n
u
1
b a
du =
n
b a
(e
ban
e
aan
) (5)
Finally,
E(Y
0Y n
) =
n
b a
(e
ban
e
aan
)
_ _
n
(6)
which is the second part of equation (14) in our text; the rst part of (14)
is obtained by plugging equations (3) and (4) into (6).
Main formulas
.
Denition: X
t1Y t
= S
t
aS
t1
= 1 R
t1Y t
I the yearly ``dollar'' return,
or one plus the yearly rate of return.
.
Hypothesis: X
t1Y t
lognormal because log X
t1Y t
@ N( Y o
2
)
1. Properties of the yearly rate of return
.
Expected value, variance, and standard deviation of the yearly rate of
return R
t1Y t
:
E(R
t1Y t
) = e

o
2
2
1 (5)
VAR(R
t1Y t
) = VAR(X
t1Y t
) = E(X
2
t1Y t
) [E(X
t1Y t
)[
2
(6)
o(R
t1Y t
) = e

o
2
2
[e
o
2
1[
1a2
(9)
.
Probability for R
t1Y t
to be between a and b:
Prob(a ` R
t1Y t
` b) = p
log(1 b)
o
_ _
p
log(1 a)
o
_ _
(10)
.
Particular case:
Prob(R
t1Y t
` E(R
t1Y t
)) = Prob(Z ` oa2)Y Z@ N(0Y 1)
2. Properties of the horizon-n rate of return R
0, n
R
0Y n
=
S
n
S
0
_ _
1an
1
.
Expected value, variance, and standard deviation of the horizon rate of
return R
0Y n
:
292 Chapter 13
E(R
0Y n
) = e

o
2
2n
1 (16)
VAR(R
0Y n
) = e
2
o
2
n
(e
o
2
n
1) (21)
o(R
0Y n
) = e

o
2
2n
(e
o
2
n
1)
1a2
(22)
.
Formulas of and o
2
in terms of E IE(X
0Y 1
) I1 R
0Y n
and
V IVAR(X
0Y 1
) = VAR(R
0Y 1
):
= log E
1
2
log 1
V
E
2
_ _
(23)
o
2
= log 1
V
E
2
_ _
(24)
E(R
0Y n
) = E (1 VaE
2
)
1
2
(
1
n
1)
1 (25)
VAR(R
0Y n
) = E
2
(1 VaE
2
)
(
1
n
1)
[(1 VaE
2
)
1
n
1[ (26)
o(R
0Y n
) = [1 E(R
0Y n
)[[(1 VaE
2
)
1
n
1[
1a2
(28)
.
Probability for R
0Y n
to be between a and b:
Prob(a ` R
0Y n
` b) = p
log(1 b)
oa

n
_ _
p
log(1 a)
oa

n
_ _
(30)
Questions
13.1. This rst exercise is for the reader who is not absolutely sure that
the moment-generating function of a normal variable log X
t1Y t
IU I
N( Y o
2
) is
U
(z) = e
z z
2
o
2
a2
, a result of considerable importance in this
chapter and the subsequent ones. You can either turn to an introductory
probability text or do the following exercise. First write the normal vari-
able N( Y o
2
) as U = oZ, where Z is the unit normal variable. Then
apply the denition of the moment-generating function and write
U
(z) = E(e
zU
) = E(e
z zoZ
) = e
z
E(e
zoZ
)
You can see that calculating the moment-generating function of
U @ N( Y o
2
) amounts to calculating the moment-generating function of
the unit normal variable Z where zo plays the role of the parameter. This
is what you should do in this exercise. (Hint: Set zo as p, for instance, and
calculate
Z
(p) = E(e
pz
) =
_
y
y
e
pz
g(z) dz, where g(z) is the unit normal
density.)
13.2. Suppose that you estimate the expected yearly continuously com-
pounded return to be = 8% and its variance o
2
= 4%. You also suppose
that these yearly returns are normally distributed. What are the expected
yearly return (compounded once a year) and its standard deviation over a
one-year horizon?
13.3. Supposing that log(S
t
aS
t1
) is normal N( Y o
2
), what is the proba-
bility that the yearly return R
t1Y t
(compounded once a year) will be lower
than its expected value E(R
t1Y t
)? (The result you will get is quite striking
in its simplicity.)
13.4. Use the result you obtained under 13.3 with the data of 13.2 to
calculate the probability that R
t1Y t
will be smaller than E(R
t1Y t
).
13.5. What is the shape of the curve depicting the probability of
R
t1Y t
being smaller than E(R
t1Y t
)? Draw this curve in [prob(R
t1Y t
`
E(R
t1Y t
))Y o[ space; use values of o from 5% to 35% in steps of 5%.
13.6. Consider again the hypotheses and data of 13.2. What are the 10-
year horizon expected return and standard deviation, supposing that the
continuously compounded yearly returns are independent?
13.7. Assuming that the continuously compounded annual returns are
independent and normally distributed N( Y o
2
):
a. Determine the probability that the n-year horizon return will be
smaller than its expected value. As in exercise 13.3, you will be surprised
by the simplicity of the result you obtain.
b. Verify your result by setting n = 1; it should yield the result you got
in 13.3.
c. What is the limit of this result when n y?
13.8. Consider the hypotheses and data of exercise 13.2 ( = 8% and
o
2
= 4%). Use two methods to nd the innite horizon expected return on
your investment.
294 Chapter 13
14
INTRODUCING THE CONCEPT OF
DIRECTIONAL DURATION
Shylock. Give him direction for this merry bond.
Chapter outline
14.1 A brief reminder on the concept of directional derivative 295
14.2 A generalization of the Macaulay and Fisher-Weil concepts of
duration to directional duration 296
14.3 Applying directional duration 299
14.3.1 The case of a parallel displacement 302
14.3.2 The case of a nonparallel displacement 303
Overview
In the immunization problem treated in chapter 10, we showed that it was
not possible to systematically protect the investor when the term structure
was subject to any kind of variation. However, we pointed out that if we
made a forecast of the possible variation of the term structure, we could
still immunize the investor. In this chapter we will show that the concept
of duration can be generalized to any kind of variation of the term struc-
ture, and we will call this new duration concept directional duration. It
stems from the concept of directional derivative.
We will show how this may help generalize the concepts of the Mac-
aulay or Fisher-Weil durations. First, we begin with a reminder of a
simple way to dene the directional derivative.
14.1 A brief reminder on the concept of directional derivative
Suppose that y = f (x
1
Y F F F Y x
j
Y F F F Y x
n
) is a dierentiable function of n
variables. Its total dierential is
dy =
n
j=1
f
x
j
(x
1
Y F F F Y x
n
) dx
j
(1)
Now suppose that each dierential dx
j
is written in the following
form:
dx
j
= a
j
dz ( j = 1Y F F F Y n) (2)
where dz is a given arbitrary small increment dierent from zero. The n
a
j
elements form an n-vector, which we may call a directional vector,
written as a. The directional derivative may be dened by substituting (2)
into (1) and dividing the result by dz:
dy
dz
=
n
j=1
f
x
j
(x
1
Y F F F Y x
j
Y F F F Y x
n
)a
i
= i a (3)
where idesignates the gradient vector of f at point (x
1
Y F F F Y x
n
). Thus the
directional derivative is the scalar product of the function's gradient i
and the directional vector a. From a geometric point of view, the
directional derivative is the slope of the tangent plane (if n = 2) or the
hyperplane (if n b 2) in the direction of the vector a at point (x
1
Y F F F Y x
n
)
in (x
1
Y F F F Y x
n
Y y) space.
14.2 A generalization of the Macaulay and Fisher-Weil concepts of duration to
directional duration
In order to keep notation simple, we will call the continuously compounded
spot interest pertaining to period [0Y t[ r
0Y t
(previously denoted R(0Y t));
r
0
0Y t
(t = 1Y F F F Y n) will then designate the continuously compounded ini-
tial interest rate structure (before this structure has received any shock);
and r
1
0Y t
will be the new interest rate structure (the structure after the
shock).
Using this continuously compounded spot rate of interest, a bond pay-
ing T cash ows c
t
(t = 1Y F F F Y T) has the following arbitrage-driven
equilibrium values before and after the shock, respectively:
B
0
0
= B(r
0
0Y 1
Y F F F Y r
0
0Y t
Y F F F Y r
0
0Y T
) =
T
t=1
c
t
e
r
0
0Y t
t
B
1
0
= B(r
1
0Y 1
Y F F F Y r
1
0Y t
Y F F F Y r
1
0Y T
) =
T
t=1
c
t
e
r
1
0Y t
t
(4)
296 Chapter 14
Its total dierential is, at initial point (r
0
0Y t
Y F F F Y r
0
0Y T
),
dB =
T
t=1
c
t
(t)e
r
0
0Y t
t
dr
0Y t
(5)
Let the increase received by each spot rate r
0
0Y t
(t = 1Y F F F Y T) be written
as
dr
0
0Y t
= a
t
dzY t = 1Y F F F Y T (6)
where dz is a small, normalized increment (for instance, 1% per year or,
in nancial parlance, 10 basis points per year) and a
t
is any given arbi-
trary number (which can be negative, positive, or zero). The a
t
's form a
directional vector a Ia
1
, F F F , a
t
, F F F , a
T
. We then have dr
0Y t
= a
t
dz,
which we can substitute in (5); dB thus becomes equal to
dB =
T
t=1
c
t
(t)e
r
0
0Y t
t
a
t
dz (7)
Dividing both sides of (7) by B
0
0
dz, we get
1
B
0
0
dB
dz
=
T
t=1
ta
t
c
t
e
r
0
0Y t
t
aB =
T
t=1
ta
t
w
t
(8)
where w
t
Ic
t
e
r
0
0Y t
t
aB is the share of each cash ow in the bond's value.
We will now introduce the concept of directional duration, denoted D
a
.
Let each weight w
t
, dened as
w
t
= c
t
e
r
0Y t
t
aB
0
0
be multiplied by the directional coecient a
t
pertaining to the change in
the spot rate i
t
, a
t
w
t
called directional weight and denoted w
a
t
Ia
t
w
t
.
These n weights form a T-element directional vector denoted w
a
. The
directional duration denoted D
a
is the weighted sum of the times of pay-
ment, the weights being w
a
t
, t = 1, F F F , T:
D
a
=
T
t=1
ta
t
c
t
e
r
0
0Y t
aB
0
0
=
T
t=1
tw
a
t
= t w
a
(9)
297 Introducing the Concept of Directional Duration
Thus the directional duration of a bond (or a bond portfolio) is just the
scalar product of the time payment vector and the directional weight
vector.
Notice that, generally, this directional duration is not a weighted
average, since the sum
T
t=1
a
t
w
t
will not usually be unitary. But this
possibility should not be excluded, because one could still observe that
there is an innity of directional vectors a such that
T
t=1
a
t
w
t
= 1, or
a w = 1. One such case would be such that all directional coecients
a
t
(t = 1Y F F F Y T) are equal to one (a would be the unit vector); D
a
would
then be the Fisher-Weil duration concept, corresponding to a parallel
displacement of the term structure curve.
We notice that the concept of a weighted average can be kept if, instead
of transforming the weights w
t
by multiplying them by the directional
coecient a
t
, we transform the times of payments by considering the T
products a
t
t rather than t. It makes sense of course, since we can choose
to weight the times of payments or the cash ows to obtain the same
results. For that matter, we can see the directional coecient as equiv-
alently multiplying by a
t
(t = 1Y F F F Y T) any one of the following:
1. the times of cash ow payments t (t = 1Y F F F Y T);
2. the cash ows in current value c
t
(t = 1Y F F F Y T);
3. the cash ows in present value c
t
e
r
0
0Y t
(t = 1Y F F F Y T); or
4. the discount factors e
r
0
0Y t
(t = 1Y F F F Y T).
Thus we have the simple relationship measuring the bond's sensitivity
to any change in the term structure of interest rates:
1
B
0
0
dB
dz
ID
a
(10)
or, equivalently,
dB
B
0
0
= D
a
dz (11)
Equation (10) may be read as follows: in linear approximation, the
relative increase of the bond's value divided by the increase dz (which
298 Chapter 14
could be called the bond's relative directional derivative) is simply minus
the directional duration.
We can verify that directional duration boils down to the Fisher-Weil
concept by setting a
t
= 1 (t = 1Y F F F Y n). We get
1
B
dB
dz
=
n
i=1
tc
t
e
r
0
0Y t
t
aB
0
0
,
as usual, or the Macaulay duration by setting r
0
0Y t
= i
0
(a constant) in the
last formula.
Note also that the simplicity of these formulas stems from the fact that
we have used continuously compounded rates of interest (and, of course,
discount rates) instead of the once-per-year compounded rates (or the rates
compounded a nite number of times per year), which makes formulas
more cumbersome.
We could say that the concept of directional duration generalizes well
the classical Macaulay and the Fisher-Weil duration concepts. Its only
drawback reinforces what we have said earlier: when attempting to pro-
tect the investor against any change in the interest rate structure, one has
to make a prediction about the exact changes in all of the spot interest
ratesand this is of course where diculties begin.
14.3 Applying directional duration
Let us now show how directional duration can be applied to the mea-
surement of the sensitivity of a bond's value when the term structure of
interest rates does not receive a parallel shift. In order to give such an
example, let us come back to the original example of a nonhorizontal spot
rate structure as it appeared in chapter 2, section 2.3, table 2.5 (or in
chapter 9, section 9.2), using the same bond (coupon: 7%; par value: 1000;
maturity: 10 years).
Our rst task is to convert these rates i
t
, which are compounded once
per year, into equivalent continuously compounded rates r
0Y t
= ln(1 i
t
)
(because (1 i
t
)
t
= e
ln(1i
t
)t
Ie
r
0
0Y t
t
). This is done in the third column of
table 14.1a. In our example, we have then considered dz = 0X1% (or 10
basis points). In the rst part of the table (part a), we take up the case of a
parallel displacement: all directional coecients a
t
are equal to one, so
dr
0
0Y t
= dz = 0X1% (or 10 basis points; t = 1Y F F F Y 10). The initial value of
the bond is denoted as B
0
0
and is equal to
B
0
0
=
T
t=1
c
t
e
r
0
0Y t
t
= 1118X254
T
a
b
l
e
1
4
.
1
P
a
r
a
l
l
e
l
a
n
d
n
o
n
p
a
r
a
l
l
e
l
d
i
s
p
l
a
c
e
m
e
n
t
s
o
f
t
h
e
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
c
u
r
v
e
:
a
n
a
p
p
l
i
c
a
t
i
o
n
o
f
t
h
e
c
o
n
c
e
p
t
o
f
d
i
r
e
c
t
i
o
n
a
l
d
u
r
a
t
i
o
n
D
a
(
a
)
P
a
r
a
l
l
e
l
d
i
s
p
l
a
c
e
m
e
n
t
o
f
t
h
e
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
c
u
r
v
e
T
e
r
m
o
f
p
a
y
m
e
n
t
C
a
s
h
o
w
I
n
i
t
i
a
l
s
p
o
t
r
a
t
e
(
c
o
m
-
p
o
u
n
d
e
d
o
n
c
e
a
y
e
a
r
)
I
n
i
t
i
a
l
s
p
o
t
r
a
t
e
(
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
)
I
n
i
t
i
a
l
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
C
a
l
c
u
l
a
t
i
o
n
o
f
F
i
s
h
e
r
-
W
e
i
l
d
u
r
a
t
i
o
n
t
e
r
m
s
N
o
r
m
a
l
i
z
e
d
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
i
n
c
r
e
a
s
e
N
e
w
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
N
e
w
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
t
C
t
i
0
Y
t
l
n
(
1
i
0
Y
t
)
=
r
0 0
Y
t
c
t
e
r
0 0
Y
t
t
t
c
t
e
r
0 0
Y
t
t
a
B
0
d
z
r
1 0
Y
t
=
r
0 0
Y
t
d
z
c
t
e
r
1 0
Y
t
t
1
7
0
0
.
0
4
5
0
.
0
4
4
0
1
7
6
6
.
9
8
6
0
.
0
5
9
9
0
.
0
0
1
0
.
0
4
5
0
1
7
6
6
.
9
1
9
2
7
0
0
.
0
4
7
5
0
.
0
4
6
4
0
6
6
3
.
7
9
5
0
.
1
1
4
1
0
.
0
0
1
0
.
0
4
7
4
0
6
6
3
.
6
6
8
3
7
0
0
.
0
4
9
5
0
.
0
4
8
3
1
4
6
0
.
5
5
5
0
.
1
6
2
5
0
.
0
0
1
0
.
0
4
9
3
1
4
6
0
.
3
7
4
4
7
0
0
.
0
5
1
0
.
0
4
9
7
4
2
5
7
.
3
7
0
0
.
2
0
5
2
0
.
0
0
1
0
.
0
5
0
7
4
2
5
7
.
1
4
1
5
7
0
0
.
0
5
2
0
.
0
5
0
6
9
3
5
4
.
3
2
7
0
.
2
4
2
9
0
.
0
0
1
0
.
0
5
1
6
9
3
5
4
.
0
5
6
6
7
0
0
.
0
5
3
0
.
0
5
1
6
4
3
5
1
.
3
4
9
0
.
2
7
5
5
0
.
0
0
1
0
.
0
5
2
6
4
3
5
4
.
0
4
1
7
7
0
0
.
0
5
4
0
.
0
5
2
5
9
2
4
8
.
4
4
1
0
.
3
0
3
2
0
.
0
0
1
0
.
0
5
3
5
9
2
4
8
.
1
0
3
8
7
0
0
.
0
5
4
5
0
.
0
5
3
0
6
7
4
5
.
7
8
5
0
.
3
2
7
5
0
.
0
0
1
0
.
0
5
4
0
6
7
4
5
.
4
2
0
9
7
0
0
.
0
5
5
0
.
0
5
3
5
4
1
4
3
.
2
3
4
0
.
3
4
8
0
0
.
0
0
1
0
.
0
5
4
5
4
1
4
2
.
8
4
7
1
0
1
0
7
0
0
.
0
5
5
0
.
0
5
3
5
4
1
6
2
6
.
4
1
1
5
.
6
0
1
7
0
.
0
0
1
0
.
0
5
4
5
4
1
6
2
0
.
1
7
8
B
0
=
1
1
1
8
X
2
5
4
F
i
s
h
e
r
-
W
e
i
l
d
u
r
a
t
i
o
n
:
D
F
W
=
7
X
6
4
1
y
e
a
r
s
B
1
=
1
1
0
9
X
7
4
8
300
(
b
)
N
o
n
p
a
r
a
l
l
e
l
d
i
s
p
l
a
c
e
m
e
n
t
o
f
t
h
e
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
c
u
r
v
e
T
e
r
m
o
f
p
a
y
m
e
n
t
C
a
s
h
o
w
I
n
i
t
i
a
l
s
p
o
t
r
a
t
e
(
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
)
I
n
i
t
i
a
l
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
D
i
r
e
c
t
i
o
n
a
l
c
o
e
c
i
e
n
t
s
v
e
c
t
o
r
~ a
C
a
l
c
u
l
a
t
i
o
n
o
f
d
i
r
e
c
t
i
o
n
a
l
t
e
r
m
s
I
n
c
r
e
a
s
e
i
n
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
N
e
w
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
:
N
e
w
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
s
t
C
t
l
n
(
1
i
0
Y
t
)
=
r
0 0
Y
t
c
t
e
r
0 0
Y
t
t
a
t
t
a
t
c
t
e
r
0 0
Y
t
t
a
B
0 0
a
t
d
z
r
1 0
Y
t
=
r
0 0
Y
t
a
t
d
z
c
t
e
r
1 0
Y
t
t
1
7
0
0
.
0
4
4
0
1
7
6
6
.
9
8
6
1
0
.
0
5
9
9
0
.
0
0
1
0
.
0
4
5
0
1
7
6
6
.
9
1
9
2
7
0
0
.
0
4
6
4
0
6
6
3
.
7
9
5
0
.
9
0
.
1
0
2
7
0
.
0
0
0
9
0
.
0
4
7
3
3
6
6
3
.
6
8
1
3
7
0
0
.
0
4
8
3
1
4
6
0
.
5
5
5
0
.
8
0
.
1
3
0
0
0
.
0
0
0
8
0
.
0
4
9
1
1
4
6
0
.
4
1
0
4
7
0
0
.
0
4
9
7
4
2
5
7
.
3
7
0
0
.
7
5
0
.
1
5
3
9
0
.
0
0
0
7
5
0
.
0
5
0
4
9
2
5
7
.
1
9
8
5
7
0
0
.
0
5
0
6
9
3
5
4
.
3
2
7
0
.
7
0
.
1
7
0
0
0
.
0
0
0
7
0
.
0
5
1
3
9
3
5
4
.
1
3
8
6
7
0
0
.
0
5
1
6
4
3
5
1
.
3
4
9
0
.
6
5
0
.
1
7
9
1
0
.
0
0
0
6
5
0
.
0
5
2
2
9
3
5
1
.
1
4
9
7
7
0
0
.
0
5
2
5
9
2
4
8
.
4
4
1
0
.
6
0
.
1
8
1
9
0
.
0
0
0
6
0
.
0
5
3
1
9
2
4
8
.
2
3
8
8
7
0
0
.
0
5
3
0
6
7
4
5
.
7
8
5
0
.
5
8
0
.
1
9
0
0
0
.
0
0
0
5
8
0
.
0
5
3
6
4
7
4
5
.
5
7
3
9
7
0
0
.
0
5
3
5
4
1
4
3
.
2
3
4
0
.
5
5
0
.
1
9
1
4
0
.
0
0
0
5
5
0
.
0
5
4
0
9
1
4
3
.
0
2
1
1
0
1
0
7
0
0
.
0
5
3
5
4
1
6
2
6
.
4
1
1
0
.
5
5
3
.
0
8
0
9
0
.
0
0
0
5
5
0
.
0
5
4
0
9
1
6
2
2
.
9
7
5
B
0 0
=
1
1
1
8
X
2
5
4
D
i
r
e
c
t
i
o
n
a
l
d
u
r
a
t
i
o
n
:
D
a
=
4
X
4
3
9
8
y
e
a
r
s
B
1 0
=
1
1
1
3
X
3
0
1
301
(This of course is the same value as the one we had obtained in chapter 2,
section 2.3, because we have used continuously compounded spot rates
that are equivalent to the former spot rates applied once a year.)
The Fisher-Weil duration, denoted as D
FW
, is equal to
D
FW
=
T
t=1
tc
t
e
r
0Y t
t
aB
0
= 7X641 years (12)
Let us now turn to directional duration. In table 14.1b, column 5 dis-
plays directional coecients a
t
, which play the role of giving to short-term
continuously compounded spot rates higher increases than to long ones:
notice that these directional coecients decline from a
1
= 1 to a
10
= 0X55.
Directional duration, denoted D
a
, is calculated in column 5 and is equal
to
D
a
=
T
t=1
a
t
c
t
e
r
0
0Y t
t
aB
0
0
= 4X4398 years (13)
The next column shows the actual increases in each continuously com-
pounded spot rate, a
t
dz (t = 1Y F F F Y T), which give rise to the new spot
rates r
1
0Y t
= r
0
0Y t
a
t
dz. When applied as discount rates to the bond's cash
ows, the new spot rates r
1
0Y t
yield the new value of the bond (B
1
0
) when
spot rates have sustained a nonparallel displacement (last column of table
14.1b). We have
B
1
0
=
T
t=1
c
t
e
r
1
0Y t
t
= 1113X301
Let us now see how directional duration D
a
fares when we want to
evaluate the sensitivity of the bond to a change in the spot rate structure.
14.3.1 The case of a parallel displacement
First, consider the parallel displacement of the structure. From table 14.1,
the bond's relative change in value is
hB
B
=
B
1
0
B
0
0
B
0
0
=
1109X748 1118X254
1118X254
= 0X761%
302 Chapter 14
In linear approximation, this relative change is given by minus the
Fisher-Weil duration (or the directional duration with unit directional
coecients), multiplied by dz = 0X1% per year:
dB
B
= D
FW
dz = 7X6405 years (0X1%ayear) = 0X764%
This shows a relative error, entailed by the linear approximation, of
Relative error =
(dBaB) (hBaB)
hBaB
= 0X45%
a very small error indeed, because of the small increase in the spot term
structure (10 basis points).
14.3.2 The case of a nonparallel displacement
From table 14.1b, the bond's relative change in value is
hB
B
=
B
1
0
B
0
0
B
0
0
=
1113X301 1118X254
1118X254
= 0X443%
In linear approximation, the relative change entailed by this nonparallel
displacement of the spot structure is given by minus the directional dura-
tion multiplied by dz = 0X1% per year. We have
dB
B
0
0
= D
a
This time, the relative error entailed by the linear approximation is
Relative error =
(dBaB) (hBaB)
hBaB
= 0X25%
about two times smaller than previously. This comes, of course, from
the fact that the nonparallel displacement involved increases in the spot
rates, each of which was smaller than in the parallel displacement case,
except for the rst term. (Our directional coecients started with 1 for the
rst term, and all other coecients were smaller than 1see the fth
column of table 14.1b.)
So there is no real problem in measuring a bond portfolio's sensitivity
to any interest structure change, but as we have already said in chapter
10, the problems arise when it comes to protecting the investor against
such changes. Fortunately, there are some ways out, and this is what we
will turn to in the next chapters.
Main formulas
.
Directional derivative:
dy
dz
=
n
j=1
f
x
j
(x
1
Y F F F Y x
j
Y F F F Y x
n
)a
i
= i a (3)
.
Directional duration: with w
t
= c
t
e
r
0Y t
t
aB
0
0
:
D
a
=
T
t=1
ta
t
c
t
e
r
0
0Y t
aB
0
0
=
T
t=1
tw
a
t
= t w
a
(9)
.
Bond's sensitivity to any change in the term structure of interest rates:
1
B
0
0
dB
dz
ID
a
(10)
or
dB
B
0
0
= D
a
dz (11)
Questions
14.1. When we supposeas in table 14.1a, column 4that the ve-year
horizon continuously compounded spot rate is, initially, r
0
0Y 5
= 5X069%,
what does it imply in terms of the forward rates for innitesimal trading
periods f (0Y ), for = 0 to = 5?
14.2. If we suppose, as we have done in this chapter, that the whole spot
structure receives an increase (either parallel or nonparallel), does it imply
an increase in forward rates?
14.3. What is the dimension (the measurement units) of the directional
coecients a
t
(t = 1Y F F F Y T) we have used in the denition of the direc-
tional duration?
14.4. This exercise can be considered as a project. Consider again a bond
with the same properties and the ones we retained in section 14.3 and the
304 Chapter 14
same initial structure. However, consider now the following directional
vector a:
Term t Directional coecient a
t
1 1
2 0.95
3 0.9
4 0.85
5 0.8
6 0.75
7 0.7
8 0.68
9 0.65
10 0.65
Use also the same normalized increment dz = 0X001 (10 basis points).
Determine in this case:
a. the exact relative change in the bond's price;
b. the bond's directional duration D
a
;
c. the relative change in the bond's price in linear approximation; and
d. the relative error you are introducing by using this linear
approximation.
Without doing any calculations, could you have predicted the magni-
tude of the results you have obtained under (a), (b), (c), and (d)? (This of
course should give you a good way to check that the results you obtained
have a good chance of being correct.)
15
A GENERAL IMMUNIZATION THEOREM,
AND APPLICATIONS
Aaron. . . . so must you resolve,
That what you cannot as you would achieve,
You must perforce accomplish as you may.
Titus Andronicus
Speed. But tell me true, will't be a match?
Launce. Ask my dog: if he say aye, it will! if he say no, it will;
if he shake his tail and say nothing, it will.
The Two Gentlemen of Verona
Chapter outline
15.1 Some brief methodological remarks 307
15.2 A general immunization theorem 308
15.3 Applications 312
15.4 Epilogue: a new approach to the calculus of variations 324
Overview
In the remainder of this book we will be concerned with protecting the
investor against nonparallel shifts in the term structure. This chapter will
be in the spirit of what we did in chapters 6 and 10. Here, however, we
will deal with both any initial term structure and any kind of shock
received by this structure. We will show how our previous results can be
generalized in a theorem. We will then put our theorem to the test by
giving a number of immunization applications that the reader will hope-
fully nd quite forceful.
In the next chapters we will consider a stochastic approach, dealing
with the justly celebrated Heath-Jarrow-Morton model. The necessary
methodology, the basic model, and applications will be presented in
chapters 16, 17, and 18, respectively.
15.1 Some brief methodological remarks
As we recalled in chapter 12, at the end of the seventeenth century and
during the eighteenth century, mathematicians were confronted with
novel problems that could not be tackled immediately with dierential or
integral calculus: those problems required analyzing relationships between
a function and a number, for instance. In order to distinguish those
objects from the more usual concept of functions, these relationships were
given the name of functionals. Here we have a problem very much like
one of those tricky questions posed by Jean Bernoulli to his brother and
other colleagues for their enjoyment and enlightenment. As we have seen
however, both an extension of duration to directional duration (chapter
14) and a variational approach (section 10.6 of chapter 10) seem to indi-
cate that it is not possible to tackle directly the question of immunizing an
investor against any kind of interest structure variation. However, we will
now suggest a method that will achieve what we are looking for with as
much precision as we need, and which also rests on transforming a func-
tional into a function, this time a function of several variables.1
We will state our theorem and demonstrate it in section 15.2. In section
15.3 we will apply it in a number of very diversewe could even call
them extremesituations. We hope, nally, that the reader will nd a
nice surprise in the epilogue of this chapter.
15.2 A general immunization theorem
We rst have to recall a concept that will play a central role. In chapters
11 and 12 we dened the continuously compounded rate of return over a
period T as R(T)T, equal to
_
T
0
f (u) du, where R(T) is the continuously
compounded rate of return per year and f (u) is the yearly forward rate
agreed upon at time 0 for an innitesimal trading period starting at u. We
will give the name G(T) to this function. We have
G(T) = R(T)T =
_
T
0
f (u) du (1)
Denition. We now dene the moment of order k of a bond portfolio as
the weighted average of the kth power of its times of payment, the weights
being the shares of the portfolio's cash ows in the initial portfolio value.
1 At the time of completion of this book, we were unfortunately unaware of the work by
Donald R. Chambers and Willard T. Carleton, `À Generalized Approach to Duration,''
Research in Finance, 7 (1988): 163181, where the authors introduced and developed a
``multiple factor duration model'' along similar lines, albeit with some dierences. It is to be
noted that D. R. Chambers, W. T. Carleton, and R. W. McEnally, using real data, have
tested their model for immunization purposes with impressive success. See their paper
`Ìmmunizing Default-Free Bond Portfolios with a Duration Vector,'' Journal of Financial
and Quantitative Analysis, 23, no. 1 (March 1988): 89104.
308 Chapter 15
Such a moment can therefore be written as
N
t=1
t
k
c
t
e
_
t
0
f (u) du
aB
0
=
N
t=1
t
k
c
t
e
R(t)t
aB
0
=
N
t=1
t
k
c
t
e
G(t)
aB
0
Suppose that an investor has a horizon H, and that when he buys a bond
portfolio the spot structure (or the forward structure) is given. However,
immediately after his purchase, this structure undergoes a shockor
equivalently, receives a variation. We are now ready to state our theorem.
Theorem. Suppose that the continuously compounded rate of return over
a period t, G(t) = R(t) t =
_
t
0
f (u) du is developable in a Taylor series
m
j=0
A
j
t
j
1
(m1)3
G
(m1)
(0t)t
m1
, 0 ` 0 ` 1, where A
j
= (1aj3)G
( j)
(0),
j = 0, 1, F F F , m. Let B
H
designate the horizon-H bond portfolio's value
after a change in the spot rate structure. This change intervenes immedi-
ately after the portfolio is bought. For an investor with horizon H to be
immunized against any such change in the spot interest structure, a su-
cient condition is that at the time of the purchase
(a) any moment of order j ( j = 0Y 1Y F F F Y m) of the bond portfolio is equal
to H
j
,
(b) the Hessian matrix of the second partial derivatives (
2
B
H
a A
2
j
) is
positive-denite.
Proof. Suppose we can write the McLaurin series:
G(t) = G(0) G
/
(0)t
1
23
G
//
(0)t
2

1
m3
G
(m)
(0)t
m
1
(m 1)3
G
(m1)
(0t)t
m1
Y 0 ` 0 ` 1 (2)
This implies that G(t) can be approximated with any desired accuracy by
a polynomial of the form
G(t) eA
0
A
1
t A
2
t
2
A
m
t
m
=
m
j=0
A
j
t
j
(3)
where A
j
= (1a j3)G
( j)
(0), j = 0, 1, F F F , m.
Consider a time span [1Y N[, where N designates the last time a bond or
a bond portfolio used for immunization will pay a cash ow. Let R(t)
designate the continuously compounded rate of return function and f (u)
309 A General Immunization Theorem, and Applications
the forward rate function for instantaneous lending, agreed on at the time
(0) the portfolio is bought. Using (3), the value of the bondor the bond
portfoliocan be written
B
0
=
N
t=1
c
t
e
_
t
0
f (u) du
=
N
t=1
c
t
e
R(t)t
e
N
t=1
c
t
e
(A
0
A
1
tA
m
t
m
)
(4)
Note that we have transformed a functional B
0
, depending on function
f (u), into a function of m 1 variables (A
0
Y F F F Y A
m
). This will be the key
to our demonstration.
We know already from chapter 6 that to immunize a portfolio we have
to guarantee a future value of the investment at a given horizon H
(H ` N). Let us then express the portfolio's future value at horizon H:
B
H
=
N
t=1
c
t
e
_
t
0
f (u) du
_ _
e
_
H
0
f (u) du
=
N
t=1
c
t
e
R(t)t
_ _
e
R(H)H
=
N
t=1
c
t
e
G(t)
_ _
e
G(H)
(5)
By the hypothesis of our theorem, we can then write
B
H
e
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
_ _
e
(A
0
A
1
HA
j
H
j
A
m
H
m
)
(6)
Given H, this future value of the portfolio expressed at horizon H remains a
function of the m 1 variables A
0
, A
1
, F F F , A
m
. We know that to immunize
our portfolio this future value should pass through a minimum in the m 1
dimension space (B
H
Y A
0
Y F F F Y A
m
). Following what we did already in
chapter 6, minimizing B
H
is tantamount to minimizing log B
H
. We have
log B
H
elog
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
_ _
A
0
A
1
H A
j
H
j
A
m
H
m
(7)
A rst-order condition for minimizing log B
H
is that the gradient of
log B
H
be zero at the initial point A
0
, F F F , A
m
. This implies the m 1 fol-
lowing equations:
log B
H
A
0
= 0
310 Chapter 15
log B
H
A
1
= 0
F
F
F
log B
H
A
m
= 0
(8)
Written more compactly, this rst-order condition is
log B
H
A
j
= 0, j = 0,1, F F F , m (9)
Let us now see what conditions (8) or (9) imply. Using (7), the rst of
these m 1 equations implies
log B
H
A
0
=
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
(1) 1 = 0 (10)
which leads to
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
= 1 (11)
Equation (11) is an accounting identity. It says that the shares of the pres-
ent value cash ows of the immunization portfolio add up to 1. But the
innocuous-looking equation (11) also carries another message: it has a nice
mathematical interpretation, which the reader will discover very soon.
Consider now the second equation of system (8). It leads to
log B
H
A
1
=
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
(t) H = 0 (12)
or
H =
N
t=1
tc
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
(13)
which amounts to equalizing horizon H to the Fisher-Weil duration.
Equivalently, it requires equalizing horizon H to the rst moment of the
bond.
Consider now the generic term of (9). It leads to
log B
H
A
j
=
N
t=1
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
(t
j
) H
j
= 0
or
H
j
=
N
t=1
t
j
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
B
0
(14)
Equation (14) is central to our theorem: it tells us that the j th moment of
the bond portfolio must be equal to the jth power of horizon H. Thus
conditions (9) are equivalent to
H
j
=
N
t=1
t
j
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
aB
0
Y j = 0Y 1Y F F F Y m (15)
This is part (a) of our theorem; part (b) results from usual second-
order conditions for a nonconstrained minimum of a function of several
variables.
Let us now come back for an instant to equation (11), resulting from
equalizing to zero our rst partial derivative of log B
H
with respect to A
0
.
We now recognize in (11) not only a necessary accounting condition but
the equality between H
0
(= 1) and the moment of order zero of the bond
portfolio (necessarily equal to 1, as you remember your statistics teacher
telling you when he introduced the moment-generating function). So
system (15) is perfectly general and applies to all moments of the bond,
starting with the moment of order zero.
Observe also that system (15) involves m 1 equations: if the order
of the Taylor expansion you choose is m, you will want to build up an
immunization portfolio of m 1 bonds.
15.3 Applications
We will now illustrate the power of immunization generated by this
method. Suppose that practitioners have determined, either through
spline or parametric methods, that the spot rate curve is given by
R(t) = 0X04 0X00248t 10
4
t
2
(1X5)10
6
t
3
(16)
This is our initial spot rate curve, represented in Figure 15.1. It implies
that the continuously compounded return over period t is
G(t) = tR(t) = 0X04t 0X00248t
2
10
4
t
3
(1X5)10
6
t
4
(17)
312 Chapter 15
So our function G(t) already has the form of a Taylor series expan-
sion. Alternatively, G(t) could be considered to be formed from a Taylor
expansion of an estimated spot rate curve.
The initial term structure R(t) corresponding to this G(t) function is
extremely close to that depicted as a simple illustration in gure 11.2 of
chapter 11, where it was denoted R(0Y T). In fact, it can hardly be visually
distinguished from R(0Y T) because of the smallness of the third-order
term.
Consider now a portfolio made up of four bonds aY bY c, and d. Their
characteristics are summarized in table 15.1. For simplicity we assume
they pay annual coupons.
0 5 10 15 20 25
0.035
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
Initial spot rate curve
S
p
o
t

r
a
t
e
New spot rate curve
Figure 15.1
Initial and new spot rate curves.
Table 15.1
Main features of bonds a, b, c, and d
Coupon
Maturity
(years) Par value
Bond a 4 7 100
Bond b 4.75 8 100
Bond c 7 15 100
Bond d 8 20 100
Suppose that you want to invest today a value of $100 for a length of
time of seven years. Your investment horizon is seven years, but unfortu-
nately you cannot nd some piece of an AAA zero-coupon, seven-year
maturity bond that would guarantee you an R(7) = 5X28% yield per
year (continuously compounded), equivalent to a terminal value of
100e
(0X0528)7
= 144X72 in seven years. On the other hand, could you be
guaranteed this seven-year return, or the corresponding terminal value, by
choosing an appropriate portfolio using bonds a, b, c, and d?
In other words, what should the numbers n
a
Y n
b
Y n
c
, and n
d
of these bonds
be for you to purchase them such that your $100 would be transformed
into that terminal value even if, just after you have made your investment,
the term structure shifts in a dramatic way? By dramatic way we mean,
for example, that the steep spot curve we have chosen (increasing from
4% to 6%, by 200 basis points, over 20 years of maturity) turns instantly
clockwise to become horizontal at a given level, for instance 5.5%; this
implies that the shortest spot rate moves up by 150 basis points and that
the longest spot rate moves down by 50 points immediately after you have
made your purchase. How should you constitute your bond portfolio in
order to secure a value extremely close to $144.72 in seven years?
Let us compute the moments m
0
, m
1
, m
2
, and m
3
for each bond. First,
table 15.2 presents the values of each bond as determined by the initial
spot rate structure. The calculations of moments m
0
Y m
1
Y m
2
, and m
3
under
the same structure are to be found in tables 15.3, 15.4, 15.5, and 15.6,
respectively.
We now have to build a portfolio of n
a
, n
b
, n
c
, and n
d
bonds such that
its rst four moments are equal to 1, H, H
2
, and H
3
(respectively to
D
0
= 1, D, D
2
, and D
3
). Since we have chosen a horizon of seven years,
those numbers are 1, 7, 49, and 343.
So we are led to solve the following system of four equations in the four
unknowns n
a
, n
b
, n
c
, n
d
:
n
a
B
a
0
B
0
m
a
0

n
b
B
b
0
B
0
m
b
0

n
c
B
c
0
B
0
m
c
0

n
d
B
d
0
B
0
m
d
0
= H
0
n
a
B
a
0
B
0
m
a
1

n
b
B
b
0
B
0
m
b
1

n
c
B
c
0
B
0
m
c
1

n
d
B
d
0
B
0
m
d
1
= H
1
n
a
B
a
0
B
0
m
a
2

n
b
B
b
0
B
0
m
b
2

n
c
B
c
0
B
0
m
c
2

n
d
B
d
0
B
0
m
d
2
= H
2
n
a
B
a
0
B
0
m
a
3

n
b
B
b
0
B
0
m
b
3

n
c
B
c
0
B
0
m
c
3

n
d
B
d
0
B
0
m
d
3
= H
3
(18)
314 Chapter 15
For simplicity, let us call
n Ithe (row) vector of unknowns (n
a
, n
b
, n
c
, n
d
)
mIthe square matrix of terms (1aB
0
)B
l
0
m
l
k
, k = 0, 1, 2, 3
l = a, b, c, d
Each of these terms is the moment of various orders of each of the bonds
entering the immunizing portfolio, weighted by the ratio of each bond's
value to the initial investment cost.2 Matrix m is
m =
0X921856 0X954295 1X110813 1X235471
5X710284 6X478267 11X00425 13X96598
38X07528 48X53687 138X4661 216X5303
260X1156 375X6751 1895X132 3779X024
_
_
_
_
2 These weights are just the rst line of matrix m.
Table 15.2
Valuation of bonds a, b, c, and d under the initial spot rate structure R(t) = 0X04 B
0X00248tC10
C4
t
2
B(1.5)10
C6
t
3
Discounted cash ows
Maturity
(years)
t
Initial spot
rate structure
R(t)
Bond a
c
a
t
e
R(t)t
Bond b
c
b
t
e
R(t)t
Bond c
c
c
t
e
R(t)t
Bond d
c
d
t
e
R(t)t
0 0.04
1 0.042378 3.83403 4.55291 6.709552 7.668059
2 0.044558 3.658956 4.34501 6.403173 7.317912
3 0.046549 3.478656 4.130904 6.087648 6.957312
4 0.04836 3.296471 3.91456 5.768825 6.592943
5 0.05 3.115203 3.699304 5.451605 6.230406
6 0.051477 2.93713 3.487842 5.139977 5.874259
7 0.0528 71.86511 3.282301 4.837075 5.528085
8 0.053978 68.01664 4.545265 5.194588
9 0.05502 4.266237 4.875699
10 0.055934 4.001101 4.572686
11 0.05673 3.75048 4.286263
12 0.057415 3.514599 4.016685
13 0.058 3.293366 3.763847
14 0.058492 3.086439 3.527358
15 0.058901 44.22598 3.306615
16 0.059235 3.100858
17 0.059503 2.909221
18 0.059714 2.730772
19 0.059877 2.564542
20 0.06 32.52897
Initial bond's value 92.18556 95.42947 111.0813 123.5471
h Ithe (column) vector of horizon (or duration) to the powers 0, 1, 2, 3
h = (0Y 7Y 49Y 343)
System (18) can now be written
nm = h (19)
which can easily solved as
n = m
1
h
where matrix m
1
is
m
1
=
43X62181 11X4693 0X818443 0X01877
48X7515 13X49682 1X00556 0X023675
10X24403 3X30703 0X307824 0X00877
3X29338 1X106155 0X11074 0X003599
_
_
_
_
Table 15.3
Moments of order 0 of bonds a, b, c, and d
Share of discounted cash ows in each bond's value
(Maturity)
0
t
0
Initial spot
rate structure
R(t) c
a
t
e
R(t)t
aB
a
0
c
b
t
e
R(t)t
aB
b
0
c
c
t
e
R(t)t
aB
c
0
c
d
t
e
R(t)t
aB
d
0
1 0.042378 0.04159 0.04771 0.060402 0.062066
1 0.044558 0.039691 0.045531 0.057644 0.059232
1 0.046549 0.037735 0.043288 0.054804 0.056313
1 0.04836 0.035759 0.04102 0.051933 0.053364
1 0.05 0.033793 0.038765 0.049078 0.050429
1 0.051477 0.031861 0.036549 0.046272 0.047547
1 0.0528 0.77957 0.034395 0.043545 0.044745
1 0.053978 0.712743 0.040918 0.042045
1 0.05502 0.038406 0.039464
1 0.055934 0.03602 0.037012
1 0.05673 0.033763 0.034693
1 0.057415 0.03164 0.032511
1 0.058 0.029648 0.030465
1 0.058492 0.027785 0.028551
1 0.058901 0.398141 0.026764
1 0.059235 0.025099
1 0.059503 0.023547
1 0.059714 0.022103
1 0.059877 0.020758
1 0.06 0.263292
Moments of order 0
(pure numbers)
1 1 1 1
316 Chapter 15
The solution n is then
n
a
= 2X99784
n
b
= 4X57423
n
c
= 0X82821
n
d
= 0X257722
Suppose now that at time , immediately after the purchase of portfolio
(n
a
Y n
b
Y n
c
Y n
d
), the initial spot structure that was steeply increasing turns
abruptly clockwise to become horizontal, at level 5.5% per year (con-
tinuously compounded for every maturity from t = 0 to t = 20 years).
Table 15.4
Moments of order 1 (Fisher-Weil durations) of bonds a, b, c, and d
Maturity times shares of discounted cash ows in
each bond's value
Maturity
t
Initial spot
rate structure
R(t) tc
a
t
e
R(t)t
aB
a
0
tc
b
t
e
R(t)t
aB
b
0
tc
c
t
e
R(t)t
aB
c
0
tc
d
t
e
R(t)t
aB
d
0
0 0.04
1 0.042378 0.04159 0.04771 0.060402 0.062066
2 0.044558 0.079382 0.091062 0.115288 0.118464
3 0.046549 0.113206 0.129863 0.164411 0.168939
4 0.04836 0.143036 0.164082 0.207733 0.213455
5 0.05 0.168964 0.193824 0.245388 0.252147
6 0.051477 0.191166 0.219293 0.277633 0.28528
7 0.0528 5.456991 0.240765 0.304817 0.313213
8 0.053978 5.70194 0.327347 0.336363
9 0.05502 0.345658 0.355179
10 0.055934 0.360196 0.370117
11 0.05673 0.371397 0.381627
12 0.057415 0.379679 0.390136
13 0.058 0.385427 0.396043
14 0.058492 0.388996 0.39971
15 0.058901 5.972108 0.40146
16 0.059235 0.401578
17 0.059503 0.400307
18 0.059714 0.397856
19 0.059877 0.394395
20 0.06 5.265842
Moments of order 1
(Fisher-Weil durations;
in years)
6.194337 6.788539 9.90648 11.30418
Let us call R
+
(t) the new spot structure (see gure 15.1). The new
values of each bond making up our portfolio are shown in table 15.7.
The total values of each bond and of the portfolio under the initial and
the new structure are summarized in table 15.8. From $100 at time 0, the
portfolio's value diminishes almost instantly (at time ) to $98.63686.
As to the value at horizon H = 7 years of the portfolio under the initial
structure, it was 100e
R(7)7
= 100e
(0X0528)7
= 144X72. In order to judge the
eciency of our immunization strategy, let us calculate the portfolio's
value at the same horizon. We obtain 98X63686e
R
+
(7)7
= 98X63686
(0X055)7
=
144X96.
This good result prods us to submit our strategy to even more radical
changes in the spot rate structure: from the initial change, let us make it
turn clockwise to twelve horizontal structures, from 1% to 12%. The
Table 15.5
(Maturity)
2
times shares of discounted cash ows in each bond's value
(Maturity)
0
t
0
Initial spot
rate structure
R(t) t
2
c
a
t
e
R(t)t
aB
a
0
t
2
c
b
t
e
R(t)t
aB
b
0
t
2
c
c
t
e
R(t)t
aB
c
0
t
2
c
d
t
e
R(t)t
aB
d
0
0
1 0.042378 0.04159 0.04771 0.060402 0.062066
4 0.044558 0.158765 0.182124 0.230576 0.236927
9 0.046549 0.339618 0.389588 0.493232 0.506817
16 0.04836 0.572145 0.656327 0.830934 0.853821
25 0.05 0.844819 0.96912 1.22694 1.260735
36 0.051477 1.146998 1.31576 1.665799 1.711682
49 0.0528 38.19894 1.685357 2.133722 2.192493
64 0.053978 45.61552 2.618775 2.690906
81 0.05502 3.110921 3.196608
100 0.055934 3.601956 3.701169
121 0.05673 4.085368 4.197896
144 0.057415 4.556142 4.681637
169 0.058 5.010553 5.148564
196 0.058492 5.445938 5.595941
225 0.058901 89.58162 6.021902
256 0.059235 6.42524
289 0.059503 6.805219
324 0.059714 7.1614
361 0.059877 7.493497
400 0.06 105.3168
Moments of order 2
(in years
2
)
41.30287 50.86151 124.6529 175.2614
318 Chapter 15
results are given in table 15.9. In each case the new portfolio's value at
horizon H is greater than the portfolio's value under the initial spot rate
curve.
Suppose now that the initial spot structure shifts in the following way:
from very steep, it turns less steep, but all spot rates increase (except the
last one). In other words, it turns around its end point, conserving some
concavity. The new short rate (R(0)) increases by 80 basis points, from
4% to 4.8%. Figure 15.2 describes this move. The new spot structure is
now given by
R(t) = 0X048 0X00146 (5X5)10

5
t (6)10
7
t
3
(20)
In that case, performing the necessary calculations as we have done
before, the immunization process gives us a seven-year horizon value of
$144.88 (instead of 144.72 before the spot structure change).
Table 15.6
(Maturity)
3
times shares of discounted cash ows in each bond's value
(Maturity)
0
t
0
Initial spot
rate structure
R(t) t
3
c
a
t
e
R(t)t
aB
a
0
t
3
c
b
t
e
R(t)t
aB
b
0
t
3
c
c
t
e
R(t)t
aB
c
0
t
3
c
d
t
e
R(t)t
aB
d
0
0 0.04
1 0.042378 0.04159 0.04771 0.060402 0.062066
8 0.044558 0.31753 0.364249 0.461152 0.473854
27 0.046549 1.018855 1.168763 1.479695 1.520452
64 0.04836 2.288582 2.625309 3.323734 3.415284
125 0.05 4.224093 4.845599 6.134701 6.303676
216 0.051477 6.881989 7.894561 9.994795 10.27009
343 0.0528 267.3926 11.7975 14.93605 15.34745
512 0.053978 364.9242 20.9502 21.52725
729 0.05502 27.99829 28.76947
1000 0.055934 36.01956 37.01169
1331 0.05673 44.93905 46.17685
1728 0.057415 54.67371 56.17965
2197 0.058 65.13719 66.93134
2744 0.058492 76.24313 78.34318
3375 0.058901 1343.724 90.32852
4096 0.059235 102.8038
4913 0.059503 115.6887
5832 0.059714 128.9052
6859 0.059877 142.3764
8000 0.06 2106.337
Moments of order 3
(in years
3
)
282.1652 393.6679 1706.076 3058.772
Let us now consider the following change in the initial structure: it
turns around its rst point; it still increases, but becomes atter, conserv-
ing some concavity. The end point (for t = 20 years) decreases from 6% to
5.5% (by 50 basis points). Figure 15.3 describes this move. The new term
structure is given by
R
++
(t) = 0X04 0X001712t (6X7)10
5
t
2
(9X5)10
7
t
3
(21)
and is pictured in gure 15.3. Under this new spot structure, the new
seven-year horizon of our portfolio is $144.80.
Let us try even more dramatic changes in the spot structure: we will
suppose that the new structures become either

R(t) plus 50 basis points
or R
++
minus 50 points. Those new structures are pictured in gures 15.4
and 15.5, respectively. In each case, the new seven-year horizon portfolio
Table 15.7
New bond values at time e (after spot structure shift) under new structure R*(t)
Discounted cash ows
Bond a Bond b Bond c Bond d
Maturity
New term
structure c
a
t
e
R
+
(t)t
c
b
t
e
R
+
(t)t
c
c
t
e
R
+
(t)t
c
d
t
e
R
+
(t)t
0
1 0.055 3.785941 4.495804 6.625396 7.571881
2 0.055 3.583337 4.255212 6.270839 7.166673
3 0.055 3.391575 4.027495 5.935256 6.78315
4 0.055 3.210075 3.811964 5.617632 6.42015
5 0.055 3.038288 3.607968 5.317005 6.076577
6 0.055 2.875695 3.414888 5.032466 5.75139
7 0.055 70.76687 3.232141 4.763154 5.443605
8 0.055 67.46282 4.508255 5.152291
9 0.055 4.266996 4.876567
10 0.055 4.038649 4.615598
11 0.055 3.822521 4.368595
12 0.055 3.617959 4.134811
13 0.055 3.424345 3.913537
14 0.055 3.241091 3.704105
15 0.055 46.89114 3.50588
16 0.055 3.318263
17 0.055 3.140687
18 0.055 2.972614
19 0.055 2.813535
20 0.055 35.95008
New bond values at
time
90.65178 94.30829 113.3727 127.68
320 Chapter 15
Table 15.8
Value of bond portfolio under initial and new spot rate structure, at times 0, e, and H
Optimal number of bonds
n
a
2.99784
n
b
4.57423
n
c
0.82821
n
d
0.257722
Value of each bond at time 0,
under initial structure
92.18556 95.42947 111.0813 123.5471
Value of each bond's share in
portfolio
276.358 436.5163 91.999 31.8408
Total F100
Value of each bond at time ,
under new structure
90.65178 94.30829 113.3727 127.68
Value of each bond's share in
portfolio
271.76 431.3878 93.8969 32.90593
Total F98X63686
Value of bond portfolio at horizon H = 7 years
(a) under initial term structure: 100e
(0X0528)7
= 144X72
(b) under new term structure: 98X63686e
(0X055)7
= 144X96
Table 15.9
Horizon HF7 years portfolio value under new spot rate structures.
Initial portfolio value before spot structure change: $144.7156
New spot rate structure
New bond portfolio's value
at horizon H = 7 years (in $)
0.01 144.7241
0.02 144.7448
0.03 144.791
0.04 144.8524
0.05 145.9218
0.06 145.9952
0.07 145.0708
0.08 145.149
0.09 145.2321
0.10 145.3237
0.11 145.4287
0.12 145.5533
0 5 10 15 20 25
0.035
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
S
p
o
t

r
a
t
e
New spot rate curve
Figure 15.2
Rotation of initial spot rate around its terminal point.
0 5 10 15 20 25
0.035
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
S
p
o
t

r
a
t
e New spot rate curve
Figure 15.3
Rotation of spot rate curve around its initial point.
322 Chapter 15
0 5 10 15 20 25
0.035
0.07
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
S
p
o
t

r
a
t
e
New spot rate curve
Figure 15.4
Nonparallel increase of spot rate curve.
0 5 10 15 20 25
0.035
0.03
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
S
p
o
t

r
a
t
e
New spot rate curve
Figure 15.5
Nonparallel decrease of spot rate curve.
value becomes $144.89 and $144.79, respectively. Once more the investor
is well protected.
At this point, the reader may wonder whether we will dare to make the
bold move to go from a steeply increasing curve to a decreasing one. Be-
fore proceeding, we looked for reassurance from Launce's dog. Since he
was not only wagging his tail but also saying ``Yes, sir!'' we moved ahead.
The new spot structure, denoted R
=
(t), is
R
=
(t) = 0X052 0X0004t (1X7)10
6
t
2
(2)10
7
t
3
(22)
This time, the short rate moves up by 120 basis points, and the 20-year
rate moves down by 150 points! (See gure 15.6.) The new seven-year
horizon value in this case is 144.86. Even Launce might be surprised that
in each case we get even more than a match.
15.4 Epilogue: a new approach to the calculus of variations
We may wonder whether the method we proposed in this chapter does not
open new perspectives on the calculus of variations. Those who have been
0 5 10 15 20 25
0.035
0.065
0.06
0.055
0.05
0.045
0.04
Maturity (years)
S
p
o
t

r
a
t
e
New spot rate curve
Figure 15.6
Change from an increasing spot curve to a decreasing one.
324 Chapter 15
studying even briey this beautiful part of calculus know that it is only in
very simple cases that variational problems possess analytical solutions.
The diculty stems from the usual complexity of the dierential equa-
tions resulting from the Euler-Lagrange equation or from its extensions
(the Euler-Poisson equation and the Ostrogradski equation).
One way to solve such problems might be to suppose that the optimal
function we are looking for can be developed in a Taylor series. The
integral representing the functional (such as
_
b
a
F(xY yY y
/
Y F F F Y y
(n)
) dx for
instance) would become a function of our m 1 variables A
0
, F F F , A
m
.
This function would not be hard to determine, thanks especially to the
integration software that is available today. Finding the optimal vector
(A
+
0
Y F F F Y A
+
m
) would then be an easy task.
Main formulas
Denition. The moment of order k of a bond portfolio as the weighted
average of the kth power of its times of payment, the weights being the
shares of the portfolio's cash ows in the initial portfolio value.
Such a moment can therefore be written as
N
t=1
t
k
c
t
e
_
t
0
f (u) du
aB
0
=
N
t=1
t
k
c
t
e
R(t)t
aB
0
=
N
t=1
t
k
c
t
e
G(t)
aB
0
Theorem. Suppose that the continuously compounded rate of return over
a period t, G(t) = R(t) t =
_
t
0
f (u) du is developable in a Taylor series
m
j=0
A
j
t
j
1
(m1)3
G
(m1)
(0t)t
m1
, 0 ` 0 ` 1, where A
j
= (1aj3)G
( j)
(0),
j = 0, 1, F F F , m. Let B
H
designate the horizon-H bond portfolio's value
after a change in the spot rate structure. This change intervenes immedi-
ately after the portfolio is bought. For an investor with horizon H to be
immunized against any such change in the spot interest structure, a su-
cient condition is that at the time of the purchase
(a) any moment of order j ( j = 0Y 1Y F F F Y m) of the bond portfolio is equal
to H
j
,
(b) the Hessian matrix of the second partial derivatives (
2
B
H
a A
2
j
) is
positive-denite.
This implies
H
j
=
N
t=1
t
j
c
t
e
(A
0
A
1
tA
j
t
j
A
m
t
m
)
aB
0
Y j = 0Y 1Y F F F Y m (15)
Let
n Ithe (row) vector of the optimal amounts of each bond in the
immunizing portfolio
mIthe square matrix of terms (1aB
0
)B
l
0
m
l
k
, k = 0, 1, 2, 3
l = a, b, c, d
h Ithe (column) vector of horizon H to the powers 0, 1, F F F , m 1
if there are m bonds in the optimal portfolio
n is then the solution of
nm = h
or
n = m
1
h
Questions
These exercises could be considered as a project.
Consider the initial spot rate curve as given by R(t) in this chapter.
Suppose that the spot curve is initially increasing, then takes an even more
dramatic counterclockwise movement than the one we considered. The
spot rate, represented by a third-order polynomial, decreases from 6% for
t = 0 to 4% for t = 20. Furthermore, its curve goes through points (5;
0.051) and (13; 0.043).
15.1. Determine the coecients of the third-order polynomial R
+
(t)
corresponding to the new spot structure.
15.2. Determine the optimal portfolio immunizing the investor under the
initial spot structure.
15.3. What is the seven-year horizon portfolio value corresponding to the
new structure? How does it compare to the same horizon portfolio value
under the initial structure?
326 Chapter 15
16
ARBITRAGE PRICING IN DISCRETE AND
CONTINUOUS TIME
Arcite. Let the event,
That never erring arbitrator, tell us
When we knew all ourselves, and let us follow
The backing of our chance.
The Two Noble Kinsmen
Chapter outline
16.1 The right price in a one-step model: a geometrical approach 328
16.2 Arbitrage at work for you 331
16.2.1 The case of an undervalued derivative (P
0
` V
0
) 331
16.2.2 The case of an overvalued derivative (P
0
b V
0
) 333
16.3 Some important properties of the derivative's value 336
16.4 Two fundamental properties of the Q-probability measure 338
16.5 A two-period evaluation model and extensions 339
16.6 The concept of probability measure change 344
16.7 Extension of arbitrage pricing to continuous-time processes 347
16.7.1 A statement of the Cameron-Martin-Girsanov theorem 347
16.7.2 An intuitive proof of the Cameron-Martin-Girsanov
theorem 348
16.8 An application: the Baxter-Rennie martingale evaluation of a
European call 352
Overview
The law of one price tells us that in a market without frictions a given
commodity cannot have various prices at one time, because otherwise
arbitrageurs would step in and reap the absolute value of the dierence
between the arbitrage price and any other price. Our purpose in this
chapter is to show how a derivative's arbitrage price can easily be estab-
lished geometrically in the so-called binomial conguration.
We suppose that at time 0 a random variable (the price of a stock, for
instance) is known; call it S
0
. At time ht, there are two states of the
nature: an up-state and a down-state, where the stock can take values S
u
and S
d
, respectively. We suppose that at that time a derivative value
corresponds to each of those values, denoted f (S
u
) and f (S
d
). There
are probabilities p and 1 p that the down- and up-states are reached,
respectively. We suppose also that a bond with maturity ht and par
B
ht
= 1 can be either bought or sold at price B
0
= e
iht
at time 0. (We
can verify that the implicit, continuously compounded interest over that
time interval is i(0Y ht) = ht
1
log(B
ht
aB
0
) = (log B
0
)ht
1
= i.1) Our
goal is now to determine the arbitrage value of the derivative at time 0.
The approach usually taken here is to surmise that building a portfolio
of a number a of stocks and a number b of bonds can replicate at time ht
the values the derivative can take. Before rushing to the corresponding
system of two equations in two unknowns, however, let us consider a geo-
metric approach. Not only will we be able to get the derivative's value
without any calculations, but we will be in a position to understand easily
all comparative statics2 of our results. There is an additional advantage in
the geometric approach: since all the results become natural, they are easy
to remember.
16.1 The right price in a one-step model: a geometrical approach
Figure 16.1 summarizes our situation at time ht. Today, S
0
= 5. After ht,
we can be either at point U or D. (To each value S
u
= 8 or S
d
= 4 corre-
sponds the derivative's value f (S
u
) = 9 or f (S
d
) = 3, respectively.) Let us
consider how we could form a portfolio such that its value at time ht
would be exactly f (S
u
) or f (S
d
) if we are in the up- or the down-state.
Consider rst a portfolio made up of a units of the stock only; a can be
positive or negative, an integer or a fraction. We understand of course
that we could then replicate all derivatives of the stock that would be
located on the ray from the origin f (S) = aS. It turns out that we have
chosen our points U and D denitely not on such a ray from the origin.
However, we notice that they are on a displaced ray from the origin.
Hence, the portfolio's value at ht must be of the form f (S) = aS ,
where is the ordinate at the origin of the line UD, equal to the amount
that must come in addition to aS to replicate both U and D. Therefore, if
b denotes the number of bonds held at time 1 and if B
1
is the value
of each bond at time 1, then = bB
1
. The coecient a is simply the
1 Notice that we can always go from prices to yields, or vice versa. Here, if i =
ht
1
log(B
ht
aB
0
), with B
ht
= 1 this last equality implies i = ht
1
(log B
0
); therefore,
log B
0
= iht and B
0
= e
iht
.
2 ``Comparative statics'' is an expression widely used in economics, dening the analysis of
changes in equilibrium values entailed by changes in parameters of the underlying model.
328 Chapter 16
V S
6 5 3 2 1 0
1
1
2
3
4
2
3
4
1 2 3 4 5 6 7
f (S)
f (S
u
)
D
U
f (S
d
)
f (S) = e
it
V
V
o
= 4.8
5
6
7
8
9
10
11
4
S
d
U
U
Q
D
8
S
u
0
b
q 1
1 q
D
S
o
e
it
7
1
Figure 16.1
A geometrical construction of a martingale probability and derivative's value in a two-period
model. The q-probability measure makes the underlying security and the derivative a mar-
tingale in present and future value. An innity of derivative contracts (D
//
U
//
, D
////
U
////
, . . . ) can
be dened with the same price V
0
and leads to the same martingale probability.
329 Arbitrage Pricing in Discrete and Continuous Time
ratio [ f (S
u
) f (S
d
)[a(S
u
S
d
); to determine b, we know that it must
be such that f (S
u
) = aS
u
bB
1
or f (S
d
) = aS
d
bB
1
. Hence, b =
B
1
1
[ f (S
u
) aS
u
[, or B
1
1
[ f (S
d
) aS
d
[ and our results: at time 0, form a
portfolio of a number a of stocks, and lend an amount bB
0
= be
iht
,
which will become = bB
1
= b a period ht later. You could, presumably,
make that loan directly, but you could also proceed as follows. Remem-
ber that there are on the market zero-coupon bonds that pay $1 at matu-
rity ht; each of them is worth e
iht
today. Therefore, you should buy a
number b of those at time 0; at time ht, you will sell them at par for
= bB
1
= b. At time 0, your portfolio is therefore aS
0
bB
0
, and this is
the value of the derivative at time 0, which we may call V
0
. We have
V
0
= aS
0
bB
0
(1)
where a, the number of stocks in the replicating strategy, is
a =
f (S
u
) f (S
d
)
S
u
S
d
(2)
and b, the number of bonds in the same strategy, is
b = B
1
1
[ f (S
u
) aS
u
[ = B
1
1
[ f (S
d
) aS
d
[ (3)
Innocuous as equations (2) and (3) may seem, they are of great impor-
tance; they mean that whatever state of nature at ht, there is one and only
one value for a and b that guarantees the portfolio to replicate the deriv-
ative. It means that a unique set of values a and b, known at time 0, cor-
responds to both states of nature at ht. Notice that the number of bonds
to be purchased, b, can be expressed solely as a function of variables that
are known at time 0:
b = B
1
0
(V
0
aS
0
) (3a)
In our example, suppose B
0
= 0X9. (This implies that if ht is one year,
the continuously compounded rate of returnor rate of interestover
that period is log 0X9 = 10X54%; equivalently, the implied rate of interest
compounded once a year would be 11.11%.) Suppose as before that
S
0
= 5; S
u
= 8; S
d
= 4; f (S
u
) = 9; f (S
d
) = 3. We should then buy
a = 1X5 shares and b = 9 (1X5)8 = 3 (1X5)4 = 3 bonds; in other
words, we should borrow the worth of 3 bonds. The cost of the portfolio
is thus (1X5)5 3(0X9) = 4X8, and that is V
0
, the price of our derivative.
At time ht, in the up-state, the portfolio is worth (1X5)8 3 = 9; in the
330 Chapter 16
down-state, it is worth (1X5)4 3 = 3, that is, it replicates exactly the
derivative's possible values at time ht.
Of course, we could also say that (2) and (3) are the solutions of the
system
aS
u
bB
1
= f (S
u
) (4)
aS
d
bB
1
= f (S
d
) (5)
but the step of solving the system in (aY b) can be skipped altogether
because the result is obvious in our geometrical approach. Before discus-
sing important properties of results (1) to (3), we have to make sure that
V
0
is indeed an arbitrage price.
16.2 Arbitrage at work for you
Let us now verify that this price is arbitrage determined; in other words,
any other price P
0
HV
0
would enable arbitrageurs (like you, of course) to
step in and prot from that situation.
16.2.1 The case of an undervalued derivative (P
0
HV
0
)
An individual A may very well make the following (erroneous) reasoning
about the value of the derivative at time 0. Suppose that the probabilities
of S
d
and S
u
are p = 0X95 and 1 p = 0X05 and that everybody, including
A and yourself, agrees that those are the true probabilities corresponding
to the event S(ht). So the expected value of the derivative at ht is
E[ f (S)[ = p f (S
d
) (1 p) f (S
u
) = 0X95(3) 0X05(9) = 3X30. (Notice also
that the expected value of the stock is 0X95(4) 0X05(8) = 4X2; it is
expected to fall from 5 by 16%.)
Individual A continues his reasoning as follows: the expected value of
the derivative at maturity is, in present value, 3X30(0X9) = 2X97. From his
old business school days, A remembers that uncertain cash ows should
be discounted with a risk-free interest rate to which an appropriate risk
premium should be added. More eager to make some quick money than
to ip through the thousand pages of his old investments text, A decides
to inict a hefty risk premium to that kind of investment (25%). The
discount rate jumps to 11X11% 25% = 36X11%, and A is then condent
that the true value of the derivative at time 0 is 3X30a1X3611 = 2X42. (If
you are interested in making a classroom experiment about the partic-
ipants' evaluations of the derivative at t = 0, this is typically the order of
magnitude you will get. We did so with 21 professionals and received an
average estimate of 2.2 for the derivative's value.)
So A thinks the derivative's value is 2.42. You come along and oer to
buy this contract for 4, a price way above his estimate and even well
above the expected value of the derivative's payos in future value (3.30).
So A eagerly accepts. He now has the obligation of paying you 3 with
probability 95%, and 9 with probability 5%, random outcomes with an
expected value of 3.30. For his eort, he receives 4 today, which he can
of course transform into 4(1X11) = 4X44 in one year with certainty. Of
course, he has diculty containing his satisfaction at having met such a
gullible counterpart.
A has, however, committed two fatal mistakes: rst, he has fallen into
the trap of considering variables that turn out to be useless (the proba-
bilities of the up- and down-states); second, he has totally neglected a
variable of central importance: today's stock value S
0
. Here now is how
you will proceed to gain 0.8 immediatelythe dierence between the true
value of the derivative (4.8) and what you have paid (4)with all your
future obligations being faithfully met.
Pay A the sum of 4 for entering the contract; you are then entitled to
the payos of the contract at its maturity. On the other hand, at time 0
you short the portfolio whose value is V
0
. (Remember it is made of a
stocks, and borrowing a future value of b.) So short-selling the portfolio
means that you take a negative position with respect to the portfolio: you
short-sell a shares of the stock and you lend a future value of b. This
amounts to receiving aS
0
and paying bB
0
(the present value of one
bond times b bonds with one-par value). In our example, you receive
(1X5)5 3(0X9) = 4X8. Your net cash ow is 4X8 4 = 0X8 at time 0.
Consider now what happens at time ht. There are two states to con-
sider. We will show that in either of those states you can exactly meet
your obligations thanks to the contract you signed with A and to the
lending you did.
.
In the up-state, you receive a payo from the derivative equal to 9; you
also receive the par of the bonds you have bought, which takes you to
12. Now you have to buy back the stocks you borrowed and remit them
to their owners. This costs you exactly (1X5)8 = 12, and you are therefore
just even.
332 Chapter 16
.
In the down-state, your payo from the contract is 3 only, to which
you can add 3 from the bonds that come to maturity. This gives you 6.
You have to buy and remit the stocks, which costs you in the down-state
(1X5)4 = 6. So your net payo is exactly 0, as in the up-state. Meanwhile,
of course, you have made at time 0 a certain net prot of 0.8 (which you
can transform into 0.888 one year later). Table 16.1 summarizes these
operations and their related cash ows.
What will happen, of course, is that A will soon realize his mistake: you
make a sure prot of 0.8. His evaluation of the contract was erroneous: he
could have sold it to you for nearly 20% more than the price he received
(4.8 instead of 4). He will thus discover that, far from being overvalued,
the price he had agreed on was grossly undervalued. The price will thus be
bid up until it reaches at time 0 the arbitrage price V
0
= aS
0
bB
0
= 4X8.
16.2.2 The case of an overvalued derivative (P
0
IV
0
)
We now have to show that if the price of the derivative is overvalued,
arbitrage will force it down to its V
0
level. Suppose that the p probabilities
are inverted and that is there is a 95% probability that up-state happens
and a 5% probability for the down-state. If B, like anybody else, takes the
expected value route, he will determine that in probability the future value
of the derivative is (0X95)9 (0X05)3 = 8X7. Not willing to take undue
risks, he also applies the hefty discount rate of 36.11% and gets a ``fair
value'' of 8X7a1X3611 = 6X39. You come along and oer to sell him this
contract for an amount of 5.5; B is thus oered to buy a contract for an
amount that is way below his estimate of the true value of the contract,
and of course he accepts.
Let us now see what you will do. You know that the derivative is
overvalued (5.5 against an arbitrage value of 4.8). So the portfolio reect-
ing the derivative is undervalued: you will then buy it. You buy a units of
the stock and b bonds, for a cost V
0
= aS
0
bB
0
= 4X8. You sell the
contract for 5.5 and prot 0.7.
At maturity, consider rst the up-state. You have to pay 9 to B and
reimburse 3, for a total of 12. On the other hand, your stocks are worth
(1X5)8 = 12. It clicks again. In the down-state, you have to pay 3 to B and
reimburse 3; but you have (1X5)4 = 6 worth of stocks to cover exactly
those costs. Table 16.2 summarizes these operations and the correspond-
ing cash ows. So you can again prot from a price not driven by arbi-
T
a
b
l
e
1
6
.
1
A
r
b
i
t
r
a
g
e
r
e
s
u
l
t
i
n
g
f
r
o
m
a
n
u
n
d
e
r
v
a
l
u
e
d
d
e
r
i
v
a
t
i
v
e
(
P
0
H
V
0
)
T
i
m
e
0
T
i
m
e
h
t
O
p
e
r
a
t
i
o
n
C
a
s
h
o
w
O
p
e
r
a
t
i
o
n
C
a
s
h
o
w
B
u
y
d
e
r
i
v
a
t
i
v
e
a
t
P
0
=
4
S
h
o
r
t
-
s
e
l
l
p
o
r
t
f
o
l
i
o
o
f
a
s
h
a
r
e
s
a
n
d
b
b
o
n
d
s
a
S
0
b
B
0
=
(
1
X
5
)
5
3
(
0
X
9
)
=
4
X
8
P
0
=
V
0
=
a
S
0
b
B
0
=
4
X
8
1
.
I
n
u
p
-
s
t
a
t
e
:
a
.
R
e
c
e
i
v
e
p
a
y
o
o
f
d
e
r
i
v
a
t
i
v
e
:
f
(
S
u
)
=
9
b
.
R
e
c
e
i
v
e
p
a
r
f
o
r
b
b
o
n
d
s
(
b
`
0
)
X
b
=
3
c
.
B
u
y
a
n
d
r
e
m
i
t
a
s
h
a
r
e
s
f
o
r
a
S
u
=
(
1
X
5
)
8
=
1
2
f
(
S
u
)
=
b
=
a
S
u
=
1
2
N
e
t
c
a
s
h
o
w
f
(
S
u
)
a
S
u
=
0
2
.
I
n
d
o
w
n
-
s
t
a
t
e
:
a
.
R
e
c
e
i
v
e
p
a
y
o
o
f
d
e
r
i
v
a
t
i
v
e
:
f
(
S
d
)
=
3
b
.
R
e
c
e
i
v
e
p
a
r
f
o
r
b
b
o
n
d
s
:
b
=
3
c
.
B
u
y
a
n
d
r
e
m
i
t
a
s
h
a
r
e
s
f
o
r
a
S
d
(
1
X
5
)
4
=
6
f
(
S
d
)
=
b
=
a
S
u
=
6
N
e
t
c
a
s
h
o
w
V
0
P
0
=
0
X
8
N
e
t
c
a
s
h
o
w
f
(
S
d
)
a
S
d
=
0
334
T
a
b
l
e
1
6
.
2
A
r
b
i
t
r
a
g
e
r
e
s
u
l
t
i
n
g
f
r
o
m
o
v
e
r
v
a
l
u
e
d
d
e
r
i
v
a
t
i
v
e
(
P
0
I
V
0
)
T
i
m
e
0
T
i
m
e
h
t
O
p
e
r
a
t
i
o
n
C
a
s
h
o
w
O
p
e
r
a
t
i
o
n
C
a
s
h
o
w
S
e
l
l
d
e
r
i
v
a
t
i
v
e
a
t
P
0
=
5
X
5
B
u
y
p
o
r
t
f
o
l
i
o
o
f
a
s
h
a
r
e
s
a
n
d
b
b
o
n
d
s
(
b
o
r
r
o
w
t
h
e
p
r
e
s
e
n
t
v
a
l
u
e
o
f
b
b
o
n
d
s
)
;
p
a
y
a
S
0
b
B
0
P
0
=
5
X
5
a
S
0
b
B
0
=
4
X
8
1
.
I
n
u
p
-
s
t
a
t
e
:
a
.
P
a
y
o
u
t
d
e
r
i
v
a
t
i
v
e
'
s
l
i
a
b
i
l
i
t
y
:
f
(
S
u
)
=
9
b
.
P
a
y
b
a
c
k
l
o
a
n
:
b
=
3
c
.
S
e
l
l
a
s
h
a
r
e
s
f
o
r
a
S
u
=
(
1
X
5
)
8
=
1
2
f
(
S
u
)
=
9
b
=
a
S
u
=
1
2
N
e
t
c
a
s
h
o
w
a
S
u
f
(
S
u
)
b
=
0
2
.
I
n
d
o
w
n
-
s
t
a
t
e
:
a
.
P
a
y
o
u
t
d
e
r
i
v
a
t
i
v
e
'
s
l
i
a
b
i
l
i
t
y
:
f
(
S
d
)
=
3
b
.
R
e
i
m
b
u
r
s
e
l
o
a
n
:
b
=
3
c
.
S
e
l
l
s
h
a
r
e
s
f
o
r
a
S
d
=
(
1
X
5
)
4
=
6
f
(
S
d
)
=
3
b
=
a
S
d
=
6
N
e
t
c
a
s
h
o
w
P
0
V
0
=
0
X
8
N
e
t
c
a
s
h
o
w
a
S
d
f
(
S
d
)
b
=
0
335
trage considerations. The market will certainly take notice and bid down
the price of that contract until it reaches a level that now may be duly
called an arbitrage equilibrium level: V
0
= aS
0
bB
0
(4.8 in our case).
16.3 Some important properties of the derivative's value
Making use of equations (1), (2), and (3), the arbitrage portfolio's (or the
derivative's) value turns out to be
V
0
=
f (S
u
) f (S
d
)
S
u
S
d
S
0
B
1
1
f (S
u
) S
u
f (S
u
) f (S
d
)
S
u
S
d
_ _
B
0
(6)
Since B
1
= B
0
e
iht
, it is also equal to
V
0
=
f (S
u
) f (S
d
)
S
u
S
d
S
0
e
iht
f (S
u
) S
u
f (S
u
) f (S
d
)
S
u
S
d
_ _
(6a)
Notice that this expression makes a lot of sense: the derivative's value is
the present value of the stocks in the portfolio, plus the present value of
the additional amount you need at maturity to replicate the derivative's
payos (that is, the present value of the bonds you need to buy or sell).
Let us rewrite V
0
by factoring out e
iht
:
V
0
= e
iht
f (S
u
)
e
iht
S
0
S
d
S
u
S
d
_ _
f (S
d
)
S
u
e
iht
S
0
S
u
S
d
_ _ _ _
(7)
Hence V
0
is a weighted average of the derivative's payos in present
value. The weights have no relationship to the initial p and 1 p proba-
bilities; any such relationship would be mere coincidence. However, these
weights can be considered probabilities in their own right and called q and
1 q respectively. (In our example, q = 0X3
"
8 and 1 q = 0X6
"
1.)
To ensure that the q's qualify as probabilities, we need only the condi-
tion S
d
` e
iht
S
0
` S
u
, which results from arbitrage considerations. Sup-
pose that the rst inequality is not met and that S
d
b e
iht
S
0
. You would
borrow S
0
and buy the stock; ht later, you would end up with at least S
d
,
which is larger than the sum you have to reimburse. On the other hand,
suppose that S
0
e
iht
b S
u
. The initial stock's value this time is denitely
336 Chapter 16
overvalued: you could short-sell the stock, earn S
0
e
iht
on the risk-free
interest market, and buy back the security at a lower price, thus earning
unlimited arbitrage prots as before.
These probabilities are made up solely of the dierent values of the
stock (S
0
, S
u
, and S
d
) and of the bond's present value. Alternatively, we
could say that besides the stock's values, the sole determinant value of the
q is the rate of interest for the time span between time 0 and time ht.
We thus have an important rst result: the derivative's price is the
present value of the expected value of the derivative's payos under a
probability measure q that is independent of the actual probabilities and
dependent on all the values, actual and potential, of the stock, as well as
the present value of the bond.
Note that the result here of the derivative's value as a weighted average
is perfectly general; it has a nice geometric interpretation. In gure 16.1,
denote on the horizontal axis the future value of the stock S
0
e
iht
; it
necessarily lies between S
d
and S
u
as we have just shown. We can use
the (S
d
Y S
u
) segment to form an axis (0Y 1) measuring probability q =
(e
iht
S
0
S
d
)a(S
u
S
d
), as shown in gure 16.1. This probability q can be
used to determine point Q on the DU line. The height of Q is thus the
weighted average of f (S
u
) and f (S
d
) with weights q and 1 q, respec-
tively, or q f (S
u
) (1 q) f (S
d
). Indeed, that height is equal to f (S
d
)
[slope of DU in ( f (S)Y q) space[ q = f (S
d
) [ f (S
u
) f (S
d
)[q = qf (S
u
)
(1 q) f (S
d
). Now in the left part of the diagram, express the ht values
in present value through the inverse of the linear transformation e
iht
V.
You thus get V
0
= e
iht
[qf (S
u
) (1 q) f (S
d
)[, which is the derivative's
value at time 0.
From the diagram we can quickly deduce important properties, which
make a lot of sense.
First, we can immediately see that in order to construct a portfolio
equivalent to the derivative's value, you will always
.
buy shares when f (S
u
) b f (S
d
);
.
short-sell shares when f (S
u
) ` f (S
d
); and
.
take no position in shares if f (S
u
) = f (S
d
); in that case your derivative
reduces to a bond whose fair value is B
0
f (S
u
) = B
0
f (S
d
).
In order to know whether you will borrow (as in our rst example),
consider all positions of the payo line UD such that it intersects the
ordinate below the origin. For a positive slope of line, for instance, bor-
rowing will always be the rule if the payos are such that
f (S
u
)
S
u
b
f (S
d
)
S
d
. On
the contrary, lending will be the rule if the equality is reversed; there will
be no borrowing or lending, as we saw earlier, if the derivative's payos
are linearly related to the underlying security (if f (S
u
)aS
u
= f (S
d
)aS
d
),
as we noted earlier).
It is interesting to realize how, for a given set of i, S
0
, S
u
, and S
d
, we
can dene an innity of derivatives that have exactly the same arbitrage
value (this results from (7)), and we can construct those geometrically.
Pivot the straight line around point Q. You get an innity of lines with
positive, negative, and zero slopes, intersecting verticals with abscissas S
d
and S
u
at points marked (D
/
Y U
/
), (D
//
Y U
//
), and so on, which are the
payos for the new derivatives whose unique value remains xed at V
0
.
Note that it may very well be the case that one of the payos is negative;
in our diagram, (D
//
Y U
//
) corresponds to such a case.
16.4 Two fundamental properties of the Q-probability measure3
We should now stress two central properties of the q-probabilities that
have been dened here.
1. The q-probabilities have been dened in such a way as to make the
stock's process a martingale, when that process is expressed either in
present value or in future value. This is easy to see. Indeed, with
q =
S
0
e
iht
S
d
S
u
S
d
, we have
S
0
= E
Q
(S
ht
e
iht
) = qS
u
e
iht
(1 q)S
d
e
iht
(8)
and the second part of our statement comes from multiplying both sides
of (8) by e
iht
.
2. The q-probabilities make the derivative's payo s process, measured in
terms of either time 0 or time ht, a martingale. Indeed, consider (7): V
0
,
the present value of the derivative, is the Q-measure expected value of
the derivative's payos in present value. Also, multiplying (7) by e
iht
, we
nd that the derivative's value today, expressed in future value, is the
Q-measure expected value of the derivative's payos.
3 A simple way to dene a probability measure is the following: it is the collection of prob-
abilities on the set of all possible outcomes (see M. Baxter and A. Rennie, An Introduction to
Derivative Pricing (Cambridge, U.K.: Cambridge University Press, 1998), p. 223).
338 Chapter 16
We thus have a preview of a fundamental result, which still needs to be
proven for multiple period processes or continuous processes: in order to
determine a derivative's arbitrage-free value, determine a Q-probability
measure that makes the underlying asset process a martingale4 in present
value (or any period's value for that matter) and determine today's de-
rivative value as the expected value of its payos in present value, under
that Q-measure. In other words, the key issue in evaluating a derivative
will be to transform random variables under a P-probability measure into
random variables under a Q-probability measure.
16.5 A two-period evaluation model and extensions
What we have shown in a one-period model can be extended in a
straightforward way to two-period models and then further to n-period
models. This is what we will do in this section. We will rst develop a
two-period model, keeping the geometric approach we developed in the
one-period model. The extension to n-period models will then come as a
natural extension.
We will rst oer an example based on the following possible values of
the underlying security and its derivative, as illustrated in gure 16.2.
Also, we suppose that between time 1 and time 2, the simple interest is
now 10%. (This means that the simple forward interest for the second
period, as determined at period 0 and applying from time 1 to time 2, is
10%. A continuously compounded interest rate corresponds to this simple
interestdenoted for simplicity i
1
equal to log(1X1) = 9X53%.) What is
known at time 0 are
1. all values of S on the tree;
2. the derivative's values only at time 2; and
3. the interest rates i
0
and i
1
(11X
"
1% and 10%, respectively).
Our task is to construct the derivative's value along the tree. We can
proceed to evaluate the derivative on each branch geometrically, as we
have done in the one-period model. Let us draw our data corresponding
4 We recall that if Z
t
is a stochastic process such that the expected value of [Z
t
[ is nite for all
t, and if the expected value of Z
t
conditional on its time path until time s (s t) is Z
s
, then
Z
t
is dened as a martingale.
to time 2 (the possible stock prices and their corresponding derivative
values) in quadrant I5 of gure 16.3. There are four possible points: UU,
UD, DU, DD. (Notice that one of the coordinates of DD is negative, due
to the fact that if S
2
is equal to 1, the derivative's value f (S
1
) is nega-
tivehere 3X5.) Consider the rst pair of points, UU and UD. We can
immediately see that to replicate the derivative's value we will have to
buy a number of stocks a
u
and a number of bonds b
u
, which means that
we will be lending some money at time 1. (Indeed, the slope of line
UDUU is positive and its ordinate at the origin, the amount we will
have to receive at time 1, is also positive.) Applying (2), the exact amount
S
0
(f(S
0
))
S
1
(f(S
1
))
S
2
(f(S
2
))
p
u
= 0.95
q
u
= 0.38
5
(4.8)
8
(9)
4
(3)
5.5
(5.3)
p
d
= 0.05
q
d
= 0.61
p
uu
= 1/2
q
uu
= 0.46
U
UU
UD
12
(13.1)
6
(7.1)
8.8
(9.9)
p
ud
= 1/2
q
ud
= 0.53
p
du
= 2/3
q
du
= 0.85
D
DU
DD
5
(4.5)
1
(3.5)
4.4
(3.3)
p
dd
= 1/3
q
dd
= 0.15
Figure 16.2
A three-period tree carrying the underlying asset's values and the corresponding derivatives
values. For each branch, the asset's and derivative's values are expressed both in current and
present value; those values are linked by the dashed lines. For the rst period, the simple
interest rate is 11X
"
1% per period; for the second period, the interest rate is 10% per period.
5 We designate by ``quadrant I'' the quadrant in the upper right part of gure 16.3. Quad-
rants II, III, and IV run counterclockwise in the same gure.
340 Chapter 16
13.1
7.1
13
12
11
10
9
8
7
6
5
4
3
2
1
0 1 2 4 5 6 7 8 10
f (S
2
) = 1.1 V
1
V
d
=
1.1
1
E
Q
[ f (S
2
)]
d
V
u
=
1.1
1
E
Q
[ f (S
2
)]
u
E
Q
[ f (S
2
)]
d
E
Q
[ f (S
2
)]
u
V
1
V
1
V
1
V
1
13
12
11
10
9
8
7
6
5
4
3
2
1
1
2
3
4
0
0 1
1 2 3 4 6 7 8 9 10 11 12
f (S
2
)
UU
UD
S
2
1 q
u
q
u
0 1
D
U
1 q q
0 1
DD
DU
1 q
d
1 q
d
q
d
Q
u
1
Q
d
S
dd
Q
o
1
S
du
S
ud
S
uu
1.1 S
d
9
8
7
6
5
4
3
2
1
0 1 2 3 4 5 6 7 8 8 10
1.11 V
1
45
V
o
=
1.11
1
E
Q
[ f (S
1
)]
V
o
= 4.8
V
o
E
Q
[ f (S
1
)]
9
8
7
6
5
4
3
2
1
0
1 2 3 4 5 6 7 8 9 10 11 12
S
1
S
d
S
u
1.11 S
o
V
u
V
d
V
1
,

1.11 V
1
9 3 5
1.1 S
u
Figure 16.3
A geometrical construction of a Q-probability measure and a derivative's valuation (V
0
F4.8)
over a two-period time span.
of a
u
is
a
u
=
13X1 7X1
12 6
= 1
Now the amount you want to receive at time 2 is the ordinate at the
origin; it is equal to 7X1 (1)6 = 13X1 (1)12 = 1X1. Since the forward
rate of interest from time 1 to time 2 is 10%, you want to lend one at time
1; the bond's price at time 1 being one, you then buy one bond at time 1.
So your replicating portfolio is made of one stock and one bond; there-
fore, its arbitrage-enforced value at time 1 is
V
u
= (1)8 (1)1 = 9
Relying on what we showed before, we could have calculated this value
just as well by determining the q-probability pertaining to that branch of
the tree; constructing that q-probability, denoted here q
uu
, as
q
uu
=
S
u
(1X1) S
ud
S
uu
S
ud
=
8X8 6
12 6
= 0X4
"
6
and applying directly the appropriate formula, transposed from (7) to our
case, which yields
V
u
= (1X1)
1
[0X4
"
6(13X1) 0X5
"
3(7X1)[ = 9
The construction of the value V
u
is indicated in quadrant II of gure
16.3. First, the point Q
1
u
of line UDUU is obtained by the abscissa of
S
u
e
i
1
ht
= 8(1X1) = 8X8 on the S axis. As before, that point splits the line
UDUU in such a way that the ordinate of Q
1
u
becomes the Q-measure
expected value of the derivative, conditional to the fact that we are in the
up-state at time 1. We thus get q
u
f (S
uu
) (1 q
u
) f (S
ud
), of which we
have to take the present value at time 1. We do this in quadrant II by
using the reciprocal of Ve
i
1
ht
= V(1X1), and thus get V
u
on the horizontal
axis. From there this value is taken (through the 45
line in quadrant III)

to U's ordinate (9) in quadrant IV.
Of course, the same procedure may be applied to the down branch, and
the reader will not be surprised that it yields a value of the derivative at
time 1 that is V
d
= 3 (since our data for time 2 have been determined
precisely to yield the data at time 1 we used in our preceding case). In that
case, the reader can see from quadrant I that the replicating portfolio will
be made of buying 2 stocks and a debt whose value at time 2 must be 5.5.
342 Chapter 16
If 5 bonds are sold at time 1, the portfolio at that time has a value of
2(4) 5(1) = 3. We are, of course, back at our initial situation, with the
value of the derivative equal to either 9 or 3 (in the up- or down-state),
and the replicating portfolio of those values is 4.8 at time 0.
The latter calculation is illustrated by the construction of point Q
d
on
line DDDU in quadrant I, from which point D in quadrant IV can be
deduced as before. We thus get the DU line in quadrant IV that corre-
sponds exactly to our gure 16.1, from which the value of the derivative
at time 0 (V
0
= 4X8) had been calculated and which is indicated again on
the V axis of quadrant III.
We conclude that in this two-period model the arbitrage price of a
derivative dened on a claim to be received at time ht has a q-probability
expected value of two expected values, each expected value being dis-
counted by one period.
We have
V
0
= e
i
0
ht
E
Q
(V
1
) = e
i
0
ht
[qV
u
(1 q)V
d
[
with V
u
and V
d
being equal, respectively, to
V
u
= e
i
1
ht
[q
u
V
uu
(1 q
u
)V
ud
[
V
d
= e
i
1
ht
[q
d
V
du
(1 q
d
)V
dd
[
Therefore, we can write
V
0
= e
(i
0
i
1
)ht
[qq
u
V
uu
q(1q
u
)V
ud
(1q)q
d
V
du
(1q)(1q
d
)V
dd
[
= e
(i
0
i
1
)ht
E
Q
(V
2
)
Basic properties. The value of the derivative fully qualies as a dis-
counted expected value of the claim at time t. This, of course, can be
generalized to an n-period model. The essential feature of that expected
value is that it is determined under a Q-probability measure that is com-
pletely independent from the P-probability measure. The Q-measure is
that which makes both the discounted underlying security and the dis-
counted corresponding derivative a martingale. Also, as we observed in
the one-step model, we are not surprised that the Q-measure makes the
underlying security and its derivative martingales when the security and
the derivative are expressed in any year's value (so there is in fact nothing
special about the present value). For instance, 4X8(1X
"
1)(1X1) = 5X8
"
6 is the
value of the derivative at time 0 expressed in terms of the second period,
and that is just the q-measure expected value of the derivative's possible
values in the second period.
That all these properties extend to n-period models will not be a sur-
prise either. The real problem comes from certain technical restrictions
that may apply if one wants to go from an n-period model to a continuous
time model. (For these, see the excellent text by Baxter and Rennie
(1998).) But the general ideas described in this chapter will carry over to
continuous time, the framework we will be using from chapter 17 onward.
16.6 The concept of probability measure change
Consider as before a nonrecombining binomial event tree. By ``nonrecom-
bining'' we mean a tree that has the following general property: at any
node, one step up followed by one step down will not necessarily lead to
the same value as the one that would correspond to a rst step down fol-
lowed by one step up. This implies that at time 1 there are two possible
values, at time 2, 2
2
= 4 possible values (instead of 3 in a recombining
tree). Our tree is dened over a number of steps in time equal to T.
At t = 0, our process has a nonrandom initial value; at time t, it has 2
t
possible values dened at 2
t
nodes. In all, there are 1 2 2
2

2
t
2
T
= 2
T1
1 nodes on the tree. To each of these nodes
corresponds one path, and a probability that is simply the product
of all probabilities from the initial node to the node considered at time
t. All these 2
T1
1 probabilities dene a P-measure. In the same way
as we showed before, we can dene on the same set of nodes a set of
2
T1
1 risk-neutral probabilities dening a Q-measure. We know that
we have to evaluate a derivative received at time T as the expectation
of the derivative in present value under the Q-measure, that is, using the
Q-probabilities.
Now let us ask the following question: would there be a way to obtain
our result by still using the event P-probabilities or the P-measure? Our
intuition here can provide an answer: it should be possible, provided that
either the relevant random variable or the P-probability undergoes a
change. Dene, on the same tree, a process in the following way: at each
node, it is the ratio of the Q-probability to the P-probability to reach that
node. This process is well dened if the following condition is met: if the
event corresponding to any node has positive P-probability, it also has
344 Chapter 16
positive Q-probability, and the converse is true. If such a condition is met,
both probability measures are said to be equivalent.
Our process, which we can denote
t
, is such that at any time t its value
at each node is the ratio of the Q-probabilities to the P-probabilities for
that node to be reached. This process is called a discrete-time Radon-
Nikodym process, to which probabilities can be attached. Figure 16.4
illustrates such a construction. For instance, to node DU corresponds a
value
q
d
q
du
p
d
p
du
for the outcome of the Radon-Nikodym derivative; under the
P-measure, the probability for
2
to take that value is p
d
p
du
; and under
measure Q, that probability is q
d
q
du
.
For horizon T, the possible values of the Radon-Nikodym process
constitute a random variable denoted dQadP. In our T = 2 example, we
have the following random variable with its associated P-measure proba-
bilities. (The actual values of our example are in parentheses.)
It can be seen immediately that multiplying X
T
(or the claims at time
T) by the Radon-Nikodym variable at time T enables us to revert to the
P-probabilities in order to calculate the Q-measure expected value of the
claim X at T. Indeed, call X
uu
, X
ud
, X
du
, and X
dd
the possible values of
the claim at T. The claim's expected value at time 0 under Q-measure is
E
Q
(X
T
) = q
u
q
uu
X
uu
q
u
q
ud
X
ud
q
d
q
du
X
du
q
d
q
dd
X
dd
and this is equal to the expected value of
dQ
dP
X
T
under P-measure:
E
P
dQ
dP
X
T
_ _
= p
u
p
uu
q
u
q
uu
p
u
p
uu
X
uu
p
u
p
ud
q
u
q
ud
p
u
p
ud
X
ud
p
d
p
du
q
d
q
du
p
d
p
du
X
du
p
d
p
dd
q
d
q
dd
p
d
p
dd
X
dd
We can also see immediately from our construction that the Radon-
Nikodym process is a P-martingale. At any point of time s (s ` t), its
value is equal to the P-measure expected value at horizon t, conditional
on the path taken up to s (which we can call the ltration of the process
at time s, denoted F
s
). We thus have
s
= E
P
(
t
[ F
s
)
Consider rst the expected value of
1
at time 0. We have E
P
(
1
[F
0
)
= p
u
q
u
p
u
p
d
q
d
p
d
= q
u
q
d
= 1, as it should. The expected value of
2
at
time 0 is E
P
(
2
[F
0
) = p
u
p
uu
q
u
q
uu
p
u
p
uu
p
u
p
ud
q
u
q
ud
p
u
p
ud
p
d
p
du
q
d
q
du
p
d
p
du
p
d
p
dd
q
d
q
dd
p
d
p
dd
= q
u
(q
uu
q
ud
) q
d
(q
du
q
dd
) = q
u
q
d
= 1, without surprise. In the
Time 0 Time 1
Time 2
p
u
= 0.95
q
u
= 0.38
p = 1
q = 1
0
= = 1
5
[4,8]
8
[9]
U
4
[3]
D
p
d
= 0.05
q
d
= 0.61
p
uu
= 1/2
q
uu
= 0.46
UU
UD
12
[13.1]
6
[7.1]
p
u
p
uu
= 0.475
q
u
q
uu
= 0.1814
p
u
p
ud
= 0.475
q
u
q
ud
= 0.2074
p
d
p
dd
= 0.016
q
d
q
dd
= 0.0916
p
d
p
du
= 0.03
q
d
q
du
= 0.5194
p
ud
= 1/2
q
ud
= 0.53
p
du
= 2/3
q
du
= 0.85
DU
DD
5
[4.5]
1
[3.5]
p
dd
= 1/3
q
dd
= 0.15
p
q
u
= = 0.40
q
u
p
u
uu
= = 0.382
q
u
q
uu
p
u
p
uu
ud
= = 0.437
q
u
q
ud
p
u
p
ud
du
= = 15.583
q
d
q
du
p
d
p
du
dd
= = 5.5
q
d
q
dd
p
d
p
dd
d
= = 12.2
q
d
p
d
Figure 16.4
Construction of the Radon-Nikodym process x
t
for t F0, 1, 2. At each node (U, D, UU, etc.)
the rst number indicates the asset value; immediately under it, the gure in brackets is the
derivative's value. For instance, at node D, 4 is the stock's value and 3 is the derivative's
value. At each node the value taken by the Radon-Nikodym random process is indicated. (For
instance, at node D the value of the Radon-Nikodym process is 12X
"
2.)
346 Chapter 16
same way, we can determine
1
as the expected values of
2
subject to
the ltration F
1
; thus E
P
(
2
[U) = p
uu
q
u
q
uu
p
u
p
uu
p
ud
q
u
q
ud
p
u
p
ud
=
q
u
p
u
(q
uu
q
ud
) =
q
u
p
u
.
The conditional expectation of
2
at D would of course lead to q
d
ap
d
.
16.7 Extension of arbitrage pricing to continuous-time processes
The risk-neutral measure (the martingale Q-measure) transforms the ini-
tial random process into a driftless martingale process. So in order to nd
a risk-neutral measure we will have to remove any drift from the original
process. When, in continuous time, we are given a random process with
drift, our rst task will be to remove the drift by changing the measure.
We saw in the discontinuous case that the Radon-Nikodym derivative
was a random process that could be considered a multiplier of either the
initial random variable or the initial set of probabilities. Suppose now
that we want to transform a continuous Wiener process with drift into a
martingale.
Will changing the drift of the initial process be like multiplying
either the initial probability measure or the random variable by a random
variable? The answer is positive and is provided by the Cameron-Martin-
Girsanov theorem.
16.7.1 A statement of the Cameron-Martin-Girsanov theorem
The Cameron-Martin-Girsanov theorem can be stated in numerous ways
with varied levels of complexity, depending in particular on the number of
Brownian motions involved. It was rst formulated by R. H. Cameron
Table 16.3
A derivation of Q-measure probabilities using the Radon-Nikodym derivative
State of nature (nodes) at T
UU UD DU DD
Radon-Nikodym derivative values
q
u
q
uu
p
u
p
uu
q
u
q
ud
p
u
p
ud
q
d
q
du
p
d
p
du
q
d
q
dd
p
d
p
dd
(0X382) (0X437) (15X583) (5X5)
Probabilities under P-measure p
u
p
uu
p
u
p
ud
p
d
p
du
p
d
p
dd
(0X475) (0X475) (0X0
"
3) (0X01
"
6)
Probabilites under Q-measure q
u
q
uu
q
u
q
ud
q
d
q
du
q
d
q
dd
(0X1814) (0X2074) (0X519
"
4) (0X091
"
6)
and W. T. Martin in 19446 in a rather dicult setting; it was then redis-
covered, apparently independently, by I. V. Girsanov in 1960.7 Its access
is somewhat easier in the Girsanov version, although it can still take a
variety of forms. For our purposes we will use only the singleBrownian
motion version as expressed in the Baxter and Rennie text8:
If W
t
is a P-Brownian motion and
t
is an F-previsible process satisfy-
ing the boundedness condition E
P
[exp(
1
2
_
T
0

2
t
dt)[ ` y, then there exists a
measure Q such that
(a) Q is equivalent to P,
(b)
dQ
dP
= exp(
_
T
0

t
dW
t

1
2
_
T
0

2
t
dt), and
(c)
~
W
t
= W
t

_
t
0

s
ds is a Q-Brownian motion; equivalently, W
t
has drift
t
under Q.
16.7.2 An intuitive proof of the Cameron-Martin-Girsanov theorem
We will now give an intuitive proof of the theorem.9 Let (t
1
Y F F F Y t
n
)
denote a series of times belonging to the time interval [0Y T[. Let W
t
be
a Brownian motion dened on [0Y T[, with W
0
= 0. The probability for W
t
to be at time t
1
within a very short vicinity dx
1
of x
1
is given by
P(W
t
1
e dx
1
) =
1

2t
1
e
x
2
1
2t
1
dx
1
(9)
The probability that the trajectory governed by W
t
goes through the
vicinity of x
1
at time t
1
and through the vicinity of x
2
at time t
2
(t
2
b t
1
) is
given by the product of the probability of getting to the vicinity of x
1
and
the probability of W
t
1
to receive an increase hW
t
1
= W
t
2
W
t
1
. It is thus
6 R. H. Cameron and W. T. Martin, ``Transformations of Wiener Integrals under Trans-
lations,'' Annals of Mathematics, 45, no. 2 (April 1944): 386396.
7 I. V. Girsanov, `Òn Transforming a Certain Class of Stochastic Processes by Absolutely
Continuous Substitution of Measures,'' Theory of Probability and Its Applications, 5, no. 3
(1960): 285301. I am grateful to Wolfgang Stummer for mentioning to me that the theorem
had another precedent; see G. Murayama, `Òn the Transition Probability Functions of the
Markov Processes,'' Nat. Sci. Rep. Ochanomizu Univ. 5 (1954): 1020, and ``Continuous
Markov Processes and Stochastic Equations,'' Rend. Circ. Mat. Palermo Ser. 4 (1955): 48
90.
8 Baxter and Rennie, op. cit., p. 74.
9 We owe this proof to an excellent colleague of ours, who unfortunately insisted upon
anonymity.
348 Chapter 16
equal to
P(W
t
1
e dx
1
Y W
t
2
e dx
2
) =
1

2t
1
e
1
2
x
2
1
t
1
1

2(t
2
t
1
)
_ e
1
2
(x
2
x
1
)
2
t
2
t
1
dx
1
dx
2
(10)
More generally, consider the probability for W
t
to go through the vicinity
of x
1
Y x
2
Y F F F Y x
n
at times t
1
Y t
2
Y F F F Y t
n
. We have
P(W
t
1
e dx
1
Y W
2
e dx
t
2
Y F F F Y W
t
n
e dx
n
)
=
1

2t
1
e
1
2
x
2
1
t
1
1

2(t
2
t
1
)
_ e
1
2
(x
2
x
1
)
2
t
2
t
1

1

2(t
n
t
n1
)
_ e
1
2
(xnx
n1
)
2
tnt
n1
dx
1
dx
2
F F F dx
n
(11)
Consider now another probabilistic model (called Q) in which the trajec-
tory is governed by a Brownian motion with drift t, where is a con-
stant. (Later we will see what happens when is allowed to be a
deterministic function of time.) The probability Q for
~
W
t
to go through
the vicinity of x
1
Y F F F Y x
n
at times t
1
Y F F F Y t
n
is given by
Q(
~
W
t
1
e dx
1
Y
~
W
t
2
e dx
2
Y F F F Y
~
W
t
n
e dx
n
)
=
1

2t
1

1

2(t
2
t
1
)
_
1

2(t
n
t
n1
)
_ e
1
2
(x
1
t
1
)
2
t
1
e
1
2
[(x
2
t
2
)(x
1
t
1
)[
2
t
2
t
1
e
1
2
[(xn tn)(x
n1
t
n1
)[
2
tnt
n1
dx
1
dx
2
dx
n
(12)
Let us now evaluate the exponent in the exponential part of (12). It can
be written as
1
2
_
(x
1
t
1
)
2
t
1
[(x
2
t
2
) (x
1
t
1
)[
2
t
2
t
1

[(x
n
t
n
) (x
n1
t
n1
)[
2
t
n
t
n1
_
=
1
2
_
x
2
1
2x
1
t
1

2
t
2
1
t
1
(x
2
x
1
)
2
2 (x
2
x
1
)(t
2
t
1
)
2
(t
2
t
1
)
2
t
2
t
1

(x
n
x
n1
)
2
2 (x
n
x
n1
)(t
n
t
n1
)
2
(t
n
t
n1
)
2
t
n
t
n1
_
=
1
2
x
2
1
t
1
(x
2
x
1
)
2
t
2
t
1

(x
n
x
n1
)
2
t
n
t
n1
_ _
x
1
(x
2
x
1
)
(x
n
x
n1
)
1
2
[
2
t
1

2
(t
2
t
1
)
2
(t
n
t
n1
)[
(13)
The rst part of the right-hand side of (13) corresponds to the P-
probability measure (see equation (11)); it is followed by a series of terms
that all cancel out except x
n
. Finally, observe that the last bracketed
term of (13) contains only
2
t
n
. Hence we may write
Q(
~
W
t
1
e dx
1
Y F F F Y
~
W
t
n
e dx
n
) = P(W
t
1
e dx
1
Y F F F Y W
t
n
e dx
n
)e
x
n
1
2
2
t
n
(14)
We have here a formula that changes a P-probability into a Q-
probability through a quantity that depends on the realization of a ran-
dom variable (x
n
), the variable as well as terminal time t
n
. Suppose now
that we consider all possible values of time on the interval [0Y T[. We will
take up similarly all possible values of x between x
0
= 0 and x
T
. Exactly
the same analysis will apply, and we can say that probability Q to have a
given trajectory under a Brownian motion with drift is the probability
of getting the same trajectory under a Brownian motion with no drift,
multiplied by e
W
T
1
2
2
T
.
Let us now see how our analysis is modied if we want to determine the
Q-probability when changing the P-model by a function of time, which we
denote (t). In equation (12) we would change all the constants into
variables (t
1
)Y (t
2
)Y F F F Y (t
n
). The exponent of the exponential would
thus become
1
2
x
2
1
t
1
(x
2
x
1
)
2
t
2
t
1

(x
n
x
n1
)
2
t
n
t
n1
_ _
[ (t
1
)(x
1
x
0
) (t
2
)(x
2
x
1
) (t
n
)(x
n
x
n1
)[
1
2
[
2
(t
1
)(t
1
t
0
)
2
(t
2
)(t
2
t
1
)
2
(t
n
)(t
n
t
n1
)[ (15)
(with t
0
= 0 and x
0
= 0).
350 Chapter 16
Suppose nally that we consider all possible instants in the interval
[0Y T[ and, similarly, all possible values our Brownian motion can take.
Then the second bracket in (15) is the realization value of the stochastic
integral of (t)dW
t
from t = 0 to t = T, and the third bracket is the simple
integral of
2
(t). Thus, if c denotes any trajectory of the Brownian
motion, we can write
Q(c) = P(c) exp
_
T
0
(t) dW
t

1
2
_
T
0
2
(t) dt
_ _
(16)
We now have a way of translating one probability measure into another.
Inasmuch as in the discrete case we can consider ratios of probabilities
to convert one model into another, here we have a ratio that could be
expressed in its limited form as
Q(c)
P(c)
= exp
_
T
0
(t) dW
t

1
2
_
T
0
2
(t) dt
_ _
This expression is also denoted
dQ
dP
(c), which is the Radon-Nikodym
derivative. The reason for this notation comes from the way expected
values of a continuous random variable under probability density func-
tions f (x) and g(x) are calculated. Suppose that f (x) corresponds to the
P model and g(x) to the Q model; D is the domain of denition of X; call
P and Q the cumulative distributions of f (x) and g(x). We have
E
P
(X) =
_
D
xf (x) dx =
_
D
x dP
E
Q
(X) =
_
D
xg(x) dx =
_
D
x dQ
=
_
D
x
g(x)
f (x)
f (x) dx I
_
D
x(x) f (x) dx
where (x) I
g(x)
f (x)
=
dQ
dx
_
dP
dx
=
dQ
dP
.
These concepts will now be put to work in describing a model of central
importance today, which we believe to be the most recent major advance
in nance, the Heath-Jarrow-Morton model of forward interest rates,
bond prices, and their derivatives. This will be the subject of our next
chapter.
16.8 An application: the Baxter-Rennie martingale evaluation of a European call
We now illustrate the considerable importance of the martingale proba-
bility measure by showing how a central result of nance (the Black-
Scholes formula) can be obtained in a straightforward way under that
framework.
The famous Black-Scholes formula for the evaluation of a European
call was a major advance for both economics and nance. It was obtained
in 1973 through a rather dicult resolution of a second-order partial dif-
ferential equation. (In fact, Fisher Black recalled that when he saw this
equation for the rst time, he did not recognize a transformation of the
propagation of heat equation, although he had studied the latter quite
extensively.)
Applying ingeniously the concept of expectation under a Q-measure of
a martingale process, Baxter and Rennie (1998) obtained what turns out
to be, in our opinion, the simplest and at the same time the most elegant
way of nding the Black-Scholes formula. We will present it here. At
some point we will even take the liberty of simplifying their method,
choosing a shortcut to obtain the limiting mean and variance of the con-
tinuously compounded rate of return process.
The reader will recall that the basic hypothesis of Black and Scholes
is the lognormal distribution of the underlying asset: log S
t
follows a
Brownian motion log S
0
t oW
t
, where W
t
@ N(0Y t). In other words,
the stock's value can be written as
S
t
= S
0
e
toW
t
= S
0
e
to

t
Z
t
Y Z
t
@ N(0Y 1) (17)
Baxter and Rennie took the following route in order to evaluate a
claim depending on the terminal value S
T
of the stock (equal to X =
max(0Y S
T
K), where K is the exercise price of the call). They rst
looked for a discontinuous (discrete) binomial process whose continuous
limit was (17). This gave them the up or down step that S
t
follows. In
turn, they determined the Q-probability measure pertaining to those steps
and the continuous limit of the process under the Q measure, which
translated simply as another Brownian motion. Finally, they calculated
the call's value today as the present value of the expected value of the
derivative under the Q measure.
Consider the discontinuous process with P-measure
1
2
:
S
u
= S
0
e
hto

ht
(18)
352 Chapter 16
S
d
= S
0
e
hto

ht
(19)
For any branch, log(S
u
aS
0
) (the continuously compounded return over
ht) follows the binomial process
log(S
u
aS
0
) = ht o

ht
(20)
log(S
d
aS
0
) = ht o

ht
(21)
with E[log(S
ht
aS
0
)[ = ht and VAR[log(S
ht
aS
0
)[ = o
2
ht under the P-
measure. Split a time period, from 0 to t, into n equal intervals, t = nht,
and consider the distribution of S
t
:
S
t
= S
0
e
B
1
B
n
(22)
where B
i
(i = 1Y F F F Y n) is the binomial process (log S
ht
aS
0
). We then have
log(S
t
aS
0
) =

n
i=1
B
i
. Following the central limit theorem, when ht 0
(and consequently, when n y),
n
i=1
B
i
tends in law toward a normal
distribution with mean nE(B
i
) = n ht = t and variance nVAR(B
i
) =
no
2
ht = o
2
t. Hence, we have
log(S
t
aS
0
) eN( tY o
2
t) (23)
or, equivalently,
S
t
= S
0
e
t
Y
t
@ N( tY o
2
t)
= S
0
e
to

t
Z
t
Y Z
t
@ N(0Y 1) (24)
Baxter and Rennie now set out to determine the Q-martingale proba-
bility measure. We know that it is independent of the original P-measure
and that it is simply given by
q =
S
0
e
rht
S
d
S
u
S
d
=
e
rht
e
hto

ht
e
hto

ht
e
hto

ht
(25)
The only hurdle in Baxter and Rennie's demonstration lies here: what
is the limit of q when ht goes to zero? For notational simplicity, let us
denote

ht
Ih. Our purpose is to show that

lim
h0
q = lim
h0
e
rh
2
e
h
2
oh
e
h
2
oh
e
h
2
oh
e
1
2
1 h

o
2
2
r
o
_ _ _ _
Let us call N(h) and D(h) the numerator and the denominator of q.
Developing N(h) and D(h) in a Taylor series and making some cancella-
tions, we end up with
N(h) = oh r
o
2
2
_ _
h
2
terms of order h
3
or higher
D(h) = 2oh terms of order h
3
or higher
Dividing N and D by 2oh, the ratio NaD can be written as
N
D
=
1
2

o
2
2
r
2o
h terms of order h
2
or higher
1 terms of order h
2
or higher
Neglecting the terms of order h
2
and above, we get the result lim
h0
q e
1
2
(1

o
2
2
r
o
h).
We can now determine the Q-measure expected value and variance of
the binomial B
i
. Denoting a I
1
o
( o
2
a2 r) and therefore, coming
back to the initial notation h = ht, q I
1
2
(1 a

ht
), we then have
E
Q
(B
i
) = ( ht o

ht
)
1
2
(1 a

ht
)
_
( ht o

ht
)
1
2
(1 a

ht
)
_
= ht(r o
2
a2) = (r o
2
a2)
t
n
(26)
(Note that has disappeared from this expected value under the Q-
measure.)
Similarly, VAR(B
i
) can be shown to converge toward o
2
ht = o
2 t
n
.
Therefore, the sum
n
i=1
B
i
converges in law toward a normal distribution
with mean (r o
2
a2)t and variance o
2
t. We can write
n
i=1
B
i
eN[(r o
2
a2)tY o
2
t[ = (r o
2
a2)t o

t
Z
t
Y Z
t
@ N(0Y 1)
(27)
Hence S
t
under the Q-martingale measure converges toward a lognormal
process that can be written as
S
t
= S
0
e
(ro
2
a2)to

t
Z
t
Y Z
t
@ N(0Y 1) (28)
We can note here that, under this Q-martingale measure, the present
value of S
t
, S
t
e
rt
is indeed a martingale. Let us determine the expected
value of S
t
's present value:
E
Q
(e
rt
S
t
) = e
rt
E
Q
[S
0
e
(ro
2
a2)to

t
Z
t
[ (29)
354 Chapter 16
Note that the expectation on the right-hand side of (29) is simply a linear
transformation of the moment-generating function of the unit normal
distribution where o

t
plays the role of the parameter. We have

e
rt
S
0
e
(ro
2
a2)t
E
Q
[e
o

t
Z
[ = e
rt
S
0
e
(ro
2
a2)t
e
o
2
ta2
= S
0
(30)
and therefore E
Q
(e
rt
S
t
) = S
0
; so S
t
is indeed a Q-martingale; the limiting
process we have taken from binomials to a normal has preserved its
martingale character.
Applying what we know from derivative evaluation, we can now cal-
culate today's call value (C
0
) on S
T
as the present value of the claim at T
under the Q-martingale measure. If this claim at T is max(S
T
K), where
K is the exercise price of the call, we simply have to calculate
C
0
= e
rT
E
Q
[max(0Y S
T
K)[
= e
rT
E
Q
[max(0Y S
0
e
(r
1
2
o
2
)To

T
Z
T
K)[ (31)
S
T
is a strictly increasing function of the realization value of ZY z. Let
us determine the value of z, denoted z, above which max(0Y S
T
K) =
S
T
K. The equation
S
0
e
(r
1
2
o
2
)To

T
z
= K (32)
implies
z =
log(KaS
0
) (r
1
2
o
2
)T
o

T
(33)
We then have
C
0
= e
rT
_
y
z
[S
T
(z) K[ f (z) dz
= e
rT
_
y
z
S
T
(z) f (z) dz K
_
y
z
f (z) dz
_ _
Ie
rT
(I
1
KI
2
) (34)
where I
1
and I
2
designate each of the above integrals, respectively.
I
1
=
S
0

2
_
y
z
e
(ro
2
a2)To

T
zz
2
a2
dz (35)
In the exponent of e, the terms in z are part of the development of
1
2
(z o

T
)
2
; therefore, this exponent can be written as
1
2
(z o

T
)
2

rT. The integration is now straightforward:
I
1
= S
0
e
rT
1

2
_
y
z
e
1
2
(zo

T
)
2
dz (36)
Let us make the change of variable y = z o

T
; to z =
log(KaS
0
)(r
o
2
2
)T
o

T
corresponds y =
log(S
0
aK)(r
o
2
2
)T
o

T
and
I
1
= S
0
e
rT
1

2
_
y
y
e
y
2
a2
(dy) = S
0
e
rT
_
y
y
1

2
e
y
2
a2
dy
= S
0
e
rT
p(y) (37)
where p(y) is the cumulative unit normal distribution corresponding to
z = y, that is, p(y) = prob(Z ` y). The second integral is
I
2
=
_
y
z
f (z) dz =
_
z
y
f (z) dz = p(z) = p(y o

T
) (38)
and therefore the call's value is
C
0
= S
0
p(y) e
rT
Kp(y o

T
) (39)
with y =
log(S
0
aK)(ro
2
a2)T
o

T
, which is the Black-Scholes formula, obtained
through a method that involved only the calculation of a simple denite
integral.
Main formulas
By far the most important formulas of this chapter are contained in the
single-Brownian version of the Cameron-Martin-Girsanov theorem, as
expressed in Baxter and Rennie10 :
If W
t
is a P-Brownian motion and
t
is an F-previsible process satisfy-
ing the boundedness condition E
P
[exp(
1
2
_
T
0

2
t
dt)[ ` y, then there exists a
measure Q such that
(a) Q is equivalent to P,
(b)
dQ
dP
= exp(
_
T
0

t
dW
t

1
2
_
T
0

2
t
dt), and
10 Baxter and Rennie, op. cit., p. 74.
356 Chapter 16
(c)
~
W
t
= W
t

_
t
0
s
ds is a Q-Brownian motion; equivalently, W
t
has drift
t
under Q.
Questions
16.1. In section 16.3, verify the steps leading from equation (6a) to
equation (7).
16.2. What is the algebraic explanation of the fact that, for given i, S
0
, S
u
,
and S
d
values, there is an innity of derivative values f (S
d
), f (S
u
) at time
ht that lead to the same derivative value at time 0, V
0
?
16.3. Using the same i, S
0
, S
u
, and S
d
values as the ones in the text, and
assuming that f (S
u
) = 4X6 and ht = 1 year, give the value of f (S
d
) that
would lead to the same derivative value at time V
0
. (You can check your
answer with the height of point D
//
in gure 16.1.)
16.4. Suppose that, in the example of the determination of a Radon-
Nikodym process given in this chapter you want to add a third period and
therefore eight realization values to the Radon-Nikodym process. What
information would you need to do just that?
16.5. Conrm that under the Q-measure the expected value of the bino-
mial B
i
(i = 1Y F F F Y n) is indeed (r o
2
a2)tan (equation (26)).
17
THE HEATH-JARROW-MORTON MODEL
OF FORWARD INTEREST RATES, BOND
PRICES, AND DERIVATIVES
Countess. It must be an answer of most monstrous size, that must ll
all demands.
All's Well That Ends Well
Chapter outline
17.1 The one-factor model 360
17.1.1 The forward rate as a Wiener process 361
17.1.2 The current short rate as a Wiener process 361
17.1.3 The cash account as a Wiener geometric process 362
17.1.4 The zero-coupon bond in current and present values as
Wiener geometric processes 363
17.1.4.1 The zero-coupon bond in current value 363
17.1.4.2 The zero-coupon bond in present value 363
17.1.5 Finding a change of probability measure 363
17.1.6 The no-arbitrage condition 365
17.2 An important application: the case of the Vasicek volatility
reduction factor 368
Appendix 17.A.1 Determining the cash account's value B(t) 372
Appendix 17.A.2 Derivation of Ito's formula, with an application to the
solution of two important stochastic dierential equations 374
Overview
Coupon-bearing bonds are portfolios of zero-coupon bonds. Because of
the random nature of interest rates (or yields), these zero-coupon bonds,
whose prices uctuate in the opposite direction of interest rates, also
qualify as random variables; inasmuch as interest rates follow random
processes, zero-coupon bonds are transforms of such random processes.
The diculty in modeling the behavior of those zero-coupons is twofold:
rst, one should agree on a stochastic process for interest rates; second
and this is even more serious and profoundone has to recognize that if,
for instance, instantaneous forward rates are driven by a Wiener process,
then the (time-dependent) drift coecient of the process has to depend on
the volatility coecient in order to prevent arbitrage opportunities. This
last property constitutes a major discovery in nance. It was made in 1989
by David Heath, Robert Jarrow, and Andrew Morton, and published in
1992 in Econometrica.
In its most general form, the Heath-Jarrow-Morton model is quite
complex since it is an n-factor model. For our purposes we will need the
one-factor version only, which we present here in some detail. As the
reader will see, it constitutes a neat, powerful, and compelling means of
analysis.
17.1 The one-factor model1
In this chapter and in the next one, we will always use the concept of the
``forward rate'' as introduced in section 11.3 of chapter 11 under the
heading ``Forward rates for instantaneous lending and borrowing.''
We repeat the denition of those forward rates, where u is the contract
inception time and z (z b u) is the time at which the loan is made for an
innitesimal period (the time at which the borrowing starts for an inn-
itesimal period):
f (uY z) = instantaneously compounded forward rate for a loan
contracted at time u, and starting at time z, for an
innitesimal period.
We will rst suppose that the forward interest rate follows a general
Wiener process. As a particular case of that forward rate, we will then
consider the current rate2 (the forward rate when starting trading time
tends toward the current time). We will dene and evaluate the current
account (the value that $1 becomes continuously when rolled over at
the current rate). These are the basic building blocks of the model. We
1 This rst section draws heavily on chapter 5, section 3, of M. Baxter and A. Rennie, An
Introduction to Derivative Pricing (Cambridge, U.K.: Cambridge University Press, 1998), pp.
142145.
2 We prefer using here the term ``current rate'' rather than the term ``spot rate'' used in
the original HJM (1992) article, since we have reserved the term ``spot rate'' for the
rate agreed upon today for a loan starting today and maturing at T, continuously com-
pounded. We designated that spot rate as R(0Y T), and we saw that it was equal to the
average of all forward rates from 0 to T with innitesimal trading periods; R(0Y T) =
T
1
_
T
0
f (0Y t) dt.
360 Chapter 17
then turn toward the zero-coupon bond's current value under the initial
probability measure (i.e., under the initial Wiener process governing the
forward rate) and to the zero-coupon's present value under the same
probability measure. We will then be led to the two crucial steps of the
Heath-Jarrow-Morton models. The rst step is to nd a change in proba-
bility measure such that the process describing the zero-coupon bond's
present value becomes a martingale. In order to do that, using Ito 's
lemma, we will derive from the present value's process a dierential
equation; in that equation, we will make a change of variable such that
the process becomes driftless. We will then take the second step, which
will amount to showing that to prevent arbitrage there must be a denite
relationship between the volatility coecient and the drift rate; we will
show the nature of that relationship. We will then be ready to use our
valuation model, which we will apply in an all-important special case.
17.1.1 The forward rate as a Wiener process
Let t belong to the time interval [0Y T[. On this interval we can consider
the evolution of the forward interest rate f (tY T) either in dierential
form:
df (tY T) = o(tY T) dW
t
:(tY T) dt (1)
or, equivalently, in integral form:
f (tY T) = f (0Y T)
_
t
0
o(sY T) dW
s

_
t
0
:(sY T) ds (2)
where W
t
stands for a standard Brownian motion.
It is important to realize that both the drift of the forward process
:(tY T) and its volatility o(tY T) are functions of the two variables t and T.
Note also that for all forward rates f (tY T) there is only one source of
uncertainty (the Brownian motion); thus, whatever their maturity, all
forward rates will be perfectly correlated. Allowing multiple sources of
uncertainty would allow for a less than perfect correlation among the
rates. This is the case in the general Heath-Jarrow-Morton model.
17.1.2 The current short rate as a Wiener process
As a particular case of equation (2) giving the stochastic evolution of
the forward rate f (tY T), let us consider the forward rate at time t for a
361 The Heath-Jarrow-Morton Model of Forward Interest Rates and Bond Prices
trading time T tending toward t (when T t), that is, when the
trading horizon is itself t (of course when the trading period remains in-
nitesimally small). This forward rate thus becomes the current short
rate at t, which we can denote as lim
Tt
f (tY T) = f (tY t) Ir(t). We then
get
r(t) = f (0Y t)
_
t
0
o(sY t) dW
s

_
t
0
:(sY t) ds (3)
17.1.3 The cash account as a Wiener geometric process
Let B(tY t) IB(t) denote the amount on the cash account at time t,
opened at time 0, with $1: B(0Y 0) = 1. Suppose it earns the short rate r(t).
A time t, the value of that account is the random variable equal to the
geometric Wiener process:
B(t) = e
r
t
0
r(s) ds
(4)
(Heath, Jarrow, and Morton call this cash account an accumulation factor
or a money market account rolling over at r(t).)
Substituting (3) into (4), this process is
B(t) = exp
_
t
0
f (0Y s) ds
_
t
0
_
s
0
o(uY s) dW
u
ds
_
t
0
_
s
0
:(uY s) du ds
_ _
(5)
Because it will soon simplify computations, we will write the two double
integrals in (5) in a dierent form. In appendix 17.A.1 we show that their
sum can be written equivalently as
_
t
0
_
t
s
o(sY u) du dW
s

_
t
0
_
t
s
:(sY u) du ds
and therefore the value of the cash account, continuously reinvested at the
stochastic rate r(t), is equal to
B(t) = exp
_
t
0
f (0Y u) du
_
t
0
_
t
s
o(sY u) du dW
s

_
t
0
_
t
s
:(sY u) du ds
_ _
(6)
The inverse of (6), B
1
(t) = e
r
t
0
r(u) du
, will be used to determine the pres-
ent value of the bond B(tY T). Of course, B
1
(t) is nothing else than the
present value (at time 0) of $1 to be received at time t.
362 Chapter 17
17.1.4 The zero-coupon bond in current and present values as Wiener geometric
processes
17.1.4.1 The zero-coupon bond in current value
Consider now the price at t of the bond maturing at T, B(tY T). It is
equal to
B(tY T) = e
r
T
t
f (tYu) du
(7)
or, making use of the forward rate given by (2),
B(tY T) = exp
_
T
t
f (0Y u) du
_
t
0
_
T
t
o(sY u) du dW
s

_
t
0
_
T
t
:(sY u) du ds
_ _
(8)
(interverting the order of integration in the last two double integrals).
17.1.4.2 The zero-coupon bond in present value
The present value (at time 0) of B(tY T) is e
r
t
0
r(u) du
B(tY T) = B
1
(t)
B(tY T). Dividing (8) by (6) and denoting by Z(tY T) the resulting present
value of B(tY T), we have
Z(tY T) = exp
_
T
0
f (0Y u) du
_
t
0
_
T
s
o(sY u) du dW
s

_
t
0
_
T
s
:(sY u) du ds
_ _
(9)
(As the reader can notice, the calculation for the double integrals is
straightforward, thanks to the change in notation just made in (5) and
explained in appendix 17.A.1.)
17.1.5 Finding a change of probability measure
When valuing derivatives, we had seen previously that we needed a
probability measure that would make the underlying security a martin-
gale in present value. The last hurdle is thus to nd a change in proba-
bility measure such that Z(tY T) becomes a martingale. One way to
perform this is to take the dierential of the stochastic process (using Ito 's
lemmasee appendix 17.A.2, part 1) and from there look for a change in
the Brownian dierential dW such that there is no drift term left.
Let us rst determine the dierential of Z(tY T). An easy way to apply
Ito 's lemma is to consider that Z(tY T) is written in the form Z(tY T) =
e
X
t
Y where X
t
is the following Wiener process:
X
t
=
_
T
0
f (0Y u) du
_
t
0
_
T
s
o(sY u) du dW
s

_
t
0
_
T
s
:(sY u) du dt (10)
We can then write the dierential of X
t
as
dX
t
=
_
T
t
o(tY u) du dW
t

_
T
t
:(tY u) du dt Iv(tY T) dW
t

_
T
t
:(tY u) du dt
(11)
where v(tY T) I
_
T
t
o(tY u) du denotes the volatility of the X
t
process.
Applying Ito 's lemma to Z(tY T) = e
X
t
, we get
dZ(tY T) = Z(tY T) v(tY T) dW
t

_
T
t
:(tY u) du dt
v
2
2
(tY T) dt
_ _
(12)
In order to nd a new probability measure that makes Z(tY T) a mar-
tingale, we should make a change in dW
t
such that equation (12) can be
written in the form
dZ(tY T) = Z(tY T)v(tY T) d
~
W
t
(13)
Indeed, it is easy to show that in those circumstances, the solution of
(13) is of the form
Z(t) = Z
0
exp
_
t
0
v( Y T) d
~
W

1
2
_
t
0
v
2
( Y T) d
_ _
(14)
which turns out to be a martingale (see 17.A.2, part 2.2). So our rst task
is to make the necessary measure change so that equation (12) is trans-
formed into a driftless geometric Wiener motion.
What we want is a relationship between d
~
W and dW
t
such that
v(tY T) dW
t

_
T
t
:(tY u) du dt
v
2
2
(tY T) dt = v(tY T) d
~
W
t
(15)
We are thus led to
d
~
W
t
= dW
t

1
v(tY T)
_
T
t
:(tY u) du dt
v(tY T)
2
dt IdW
t

t
dt
364 Chapter 17
and hence the change of measure
t
is
t
=
1
v(tY T)
_
T
t
:(tY u) du
v(tY T)
2
(16)
17.1.6 The no-arbitrage condition
The fundamental contribution of Heath, Jarrow, and Morton was to
show that in order to avoid arbitrage, the volatility function had to be
related to the original drift term through a constraint. This important re-
sult comes from the following reasoning: in order to prevent arbitrage
opportunities, any bond with shorter maturity than T must have the same
change of measure
t
. This implies that
t
must be independent from T.
Multiply then (16) by v(tY T) and dierentiate with respect to T; we get
:(tY T) =
v(tY T)
T
[v(tY T)
t
[ = o(tY T)[v(tY T)
t
[ (17)
Equations (16) and (17) are the two major constraints of the single-
factor Heath-Jarrow-Morton model. Equation (16) stems from the change
of drift necessitated by the transformation of Z(tY T) into a martingale;
(17) results from the need to suppress arbitrage opportunities.
We thus arrive at the fundamental result of the model: the determina-
tion of the value for the drift coecient of f (tY T) under the
~
W
t
Brownian
motion. Recall that, from the basic property of the forward rate,
t
:(tY T) dt (1)
In this equation, replace dW
t
with d
~
W
t

t
dt and :(tY T) with constraint
(17). We get
df (tY T) = o(tY T)(d
~
W
t

t
dt) o(tY T) v(tY T)
t
[ [ dt
= o(tY T) d
~
W
t
o(tY T)v(tY T) dt (18)
Thus in the single-factor Heath-Jarrow-Morton model, under the mar-
tingale measure, we must have a drift coecient equal to
o(tY T)v(tY T) = o(tY T)
_
T
t
o(tY u) du
The following two-page ``guide'' summarizes the Heath-Jarrow-
Morton one-factor stochastic model of bond evaluation. We will now
apply the model in an important example.
A Guide to the Heath-Jarrow-Morton One-Factor Model: Four
Essential Steps3
1. Hypotheses
.
Evolution of forward rate f (tY T):
t
:(tY T) dt (1)
or
f (tY T) = f (0Y T)
_
t
0
o(sY T) dW
s

_
t
0
:(sY T) ds (2)
.
Particular case: evolution of current short rate:
r(t) I f (tY t) = f (0Y t)
_
t
0
o(sY t) dW
s

_
t
0
:(sY t) ds (3)
2. Valuations
.
Cash account:
B(t) = e
r
t
0
r(s) ds
= exp
_ _
t
0
f (0Y u) du
_
t
0
_
t
s
o(sY u) du dW
s
_
t
0
_
t
s
:(sY u) du ds
_
(6)
.
Zero-coupon bond in current value:
B(tY T) = e
r
T
t
f (tYu) du
= exp
_
_
T
t
f (0Y u) du
_
t
0
_
T
t
o(sY u) du dW
s
_
t
0
_
T
t
:(sY u) du ds
_
(8)
3 The numbering of the equations in this summary is that used in the text. We suppose
throughout this ``guide'' that a number of technical conditions on o(tY T) and :(tY T),
as expressed in Baxter and Rennie (op. cit., p. 143), are met. Also, for the justication
for the interchanging of limits in the double integrals involving stochastic dier-
entials, see D. Heath, R. Jarrow, and A. Morton, ``Bond Pricing and the Term
Structure of Interest Rates: A New Methodology for Contingent Claims Valuation,''
Econometrica, 60, no. 1 (January 1992), appendix.
366 Chapter 17
.
Zero-coupon bond in present value:
Z(tY T) = B
1
(t)B(tY T)
= exp
_
_
T
0
f (0Y u) du
_
t
0
_
T
s
o(sY u) du dW
s
_
t
0
_
T
s
:(sY u) du ds
_
(9)
3. Finding a change in probability measure that makes Z(t,T) a
martingale
.
Apply Ito 's lemma to Z(tY T), set v(tY T) =
_
T
t
o(tY u) du:
t

_
T
t
:(tY u) du dt
v
2
2
(tY u) dt
_ _
(12)
.
Set d
~
W
t
equal to
d
~
W
t
= dW
t

1
v(tY T)
_
T
t
:(tY u) du dt
v(tY T)
2
dt IdW
t

t
dt
with the change of measure
t
:
t
=
1
v(tY T)
_
T
t
:(tY u) du
v(tY T)
2
(16)
4. The fundamental no-arbitrage condition: a restriction on the
change of measure g
t
.
The change of measure
t
must be independent from T; set the
dierential of
t
with respect to T equal to zero to obtain
:(tY T) = o(tY T) v(tY T)
t
[ [ (17)
.
In equation (1) replace dW
t
with d
~
W
t
dt and :(tY T) with the
right-hand side of (17).
.
Then
df
t
= o(tY T) d
~
W
t
Thus the drift coecient under the martingale measure must be
_
T
t
o(tY u) duX
17.2 An important application: the case of the Vasicek volatility reduction factor
In a pioneering study of the stochastic term structure, Oldrich Vasicek4
introduced a volatility o(tY T) coecient in the form of o exp[z(T t)[,
where o and z are positive constants. We will make use of that hypothesis
in the Heath-Jarrow-Morton model. Thus we can rst calculate the value
of the drift coecient :(tY T) under the martingale measure, with no
arbitrage opportunities. Applying (17), we obtain
:(tY T) = oe
z(Tt)
_
T
t
oe
z(ut)
du = o
2
e
z(Tt)
e
2z(Tt)
z
_ _
(19)
The forward rate process, under those circumstances, is thus
df (tY T) = oe
z(Tt)
d
~
W
t

o
2
z
e
z(Tt)
e
2z(Tt)
_ _
dt (20)
Integrating (20), we can write the process governing the forward rate as
f (tY T) = f (0Y T) o
_
t
0
e
z(Tv)
d
~
W
v

o
2
z
_
t
0
e
z(Tu)
e
2z(Tu)
_ _
du
= f (0Y T) o
_
t
0
e
z(Tv)
d
~
W
v
o
2
2z
2
e
2zT
(e
2zt
1) 2e
zT
(e
zt
1)
_
(21)
From (21), it is now possible to determine both B(tY T) and r(t). First,
calculate the bond's current value B(tY T). Note that now the integrating
variable is the second argument of f (X).
B(tY T) = exp
_
T
t
f (tY u) du
_ _
= exp
_
_
T
t
f (0Y u) du
o
2
2z
2
_
T
t
[e
2zu
(e
2zt
1) 2e
zu
(e
zt
1)[ du
o
_
T
t
_
t
0
e
z(uv)
d
~
W
v
du
_
4 O. Vasicek, `Àn Equilibrium Characterization of the Term Structure,'' Journal of Financial
Economics, 5 (November 1977): 177188.
368 Chapter 17
= exp
_
_
T
t
f (0Y u) du
o
2
4z
3
[(1 e
2zt
)(e
2zT
e
2zt
)
4(e
zt
1)(e
zT
e
zt
)[ o
_
T
t
_
t
0
e
z(uv)
d
~
W
v
du
_
(22)
This is the value of B(tY T)'s bond, viewed as a geometric Wiener pro-
cess. There is unfortunately no way to express the stochastic integral
_
t
0
e
z(uv)
d
~
W
v
in closed form; however, there is a nice way out, which
was rst presented by Heath, Jarrow, and Morton in their classic 1992
paper, in the special case of z = 0 (that is, o(tY T) = o, a constant). Here
let us rst derive r(t) by writing in (21) f (tY t) = r(t):
r(t) = f (0Y t) o
_
t
0
e
z(tv)
d
~
W
v

o
2
2z
2
(2e
zt
e
2zt
1) (23)
(It can be checked that if z = 0, then r(t) = f (0Y t) o
~
W
t

o
2
t
2
2
as in
the 1992 paperby applying L'Hospital's rule to the last term of (23)
twice.) Consider now the exponential part of the last term of (22). It can
be written as
o
_
T
t
_
t
0
e
z(uv)
d
~
W
v
du = o
_
T
t
_
t
0
e
z(uttv)
d
~
W
v
du
= o
_
T
t
_
t
0
e
z(ut)
e
z(tv)
d
~
W
v
du
= o
_
t
0
e
z(tv)
d
~
W
v

e
z(Tt)
1
z
_ _
(24)
Using equation (23), the stochastic integral o
_
t
0
e
z(tv)
d
~
W
v
can be
expressed as the dierence between the Wiener process r(t) and f (0Y t)
plus a deterministic function of time:
o
_
t
0
e
z(tv)
d
~
W
v
= r(t) f (0Y t)
o
2
2z
2
(2e
zt
e
2zt
1)
and thus the term containing the double integral in (22) can be written as
o
_
T
t
_
t
0
e
z(uv)
d
~
W
v
du
= r(t) f (0Y t)
o
2
2z
2
(2e
zt
e
2zt
1)
_ _
e
z(Tt)
1
z
_ _
(25)
Using (25), we can now express the double integral in (22) in terms of
the random variable r(t). The bond's value thus becomes
B(tY T) = exp
_
_
T
t
f (0Y u) du
o
2
4z
3
[(1 e
2zt
)(e
2zT
e
2zt
)
4(e
zt
1)(e
zT
e
zt
)[
_
r(t) f (0Y t)
o
2
2z
2
(2e
zt
e
2zt
1)
_
e
z(Tt)
1
z
_ __
(26)
After performing the appropriate calculations and cancellations of the
deterministic terms in (26), we can write it as
B(tY T) = exp
_
_
T
t
f (0Y u) du [r(t) f (0Y t)[
_
e
z(Tt)
1
z
_
o
2
4z
_
1 e
2z(Tt)
2e
z(Tt)
z
2
e
2zt
_
1 e
2z(Tt)
2e
z(Tt)
z
2
___
(27)
We now introduce the following simplifying notation, which will play
an important role:
X(tY T) I
1 e
z(Tt)
z
(28)
Using this notation and observing that 1 e
2z(Tt)
2e
z(Tt)
=
1 e
z(Tt)
_ _
2
Y we can write B(tY T) as
B(tY T) = exp
_
_
T
t
f (0Y u) du [r(t) f (0Y t)[X(tY T)
o
2
4z
X
2
(tY T)(1 e
2zt
)
_
(29)
One last simplication is in order. The rst exponential term on the
right-hand side of (29) is
370 Chapter 17
e
r
T
0
f (0Yu) du
= e
[r
T
t
f (0Yu) dur
t
0
f (0Yu) du[
= e
r
T
0
f (0Yu) du
e
r
t
0
f (0Yu) du
=
B(0Y T)
B(0Y t)
(30)
The bond's value can thus be expressed as a function of the random short
rate r(t):
B(tY T) =
B(0Y T)
B(0Y t)
exp [r(t) f (0Y t)[X(tY T)
o
2
4z
X
2
(tY T)(1 e
2zt
)
_ _
(31)
This is the important formula given by Robert Jarrow and Stuart Turn-
bull in their excellent book Derivative Securities,5 where they use the fol-
lowing notation:
r(t) f (0Y t) II(t)
and
o
2
4z
X
2
(tY T)(1 e
2zt
) Ia(tY T)
At any time t, a bond with maturity T has the following value: it is the
product of a nonrandom number (the ratio of the prices at time 0 of two
bonds with maturities T and t), which is perfectly well known, and a
random variable, the exponential part of (31), which we will call J(tY T).
We thus have
B(tY T) =
B(0Y T)
B(0Y t)
J(tY T) (31a)
We will want to verify that this last (random) number collapses to one
if we take uncertainty out of the model by setting o(tY T) = 0. But rst we
should check that B(tY T) takes the right values when t = 0 and t = T.
.
At t = 0Y r(0) = f (0Y 0); I = r(0) f (0Y 0) = 0; also, a = 0. Therefore
B(0Y T) =
B(0Y T)
B(0Y 0)
= B(0Y T), as it should.
5 R. Jarrow and S. Turnbull, Derivative Securities (Cincinnati, Ohio: South-Western, 1996).
.
At t = TY X(TY T) = 0; a(TY T) = 0; thus B(TY T) = 1.
.
Suppose now that we remove all uncertainty by setting o = 0. Taking
into account the nonarbitrage restriction, the current short rate at t, r(t),
is always equal to the forward rate for trading at t, f (0Y t) and is a con-
stant, which we can denote by r. In those conditions of certainty, I = 0
and a = 0.6 Thus J(tY T) = 1 and
B(tY T) =
B(0Y T)
B(0Y t)
=
e
rT
e
rt
= e
r(Tt)
as it should.
Before going further, let us examine the exact nature of the random
variable r(t), since that variable carries all the randomness of the bond's
value. From (23) the current, short rate can be written as
r(t) = f (0Y t)
o
2
2
1 e
zt
z
_ _
2
oe
zt
_
t
0
e
zv
d
~
W
v
(32)
(We can verify that when t = 0, r(0) = f (0Y 0), as it should.)
The stochastic integral
_
t
0
e
zv
d
~
W
v
does not have a closed form; how-
ever, we know7 that it is a normal variable with mean 0 and variance
_
t
0
e
2zv
dv =
1
2z
(e
2zt
1). Thus the mean of r(t), under the martingale
measure, is f (0Y t)
o
2
2
_
1e
zt
z
_
2
and its variance is o
2
e
2zt 1
2z
(e
2zt
1) =
o
2
2z
(1 e
2zt
).
We will now give an example of application of this one-factor Heath-
Jarrow-Morton model by considering the problem of bond immunization.
This will be the subject of our next chapter.
Appendix 17.A.1 Determining the cash account's value B(t)
We explain here how to change the writing of the double integrals
in equation (5), which gives the cash account's value. We start with
6 See the denition of a(tY T) abovebefore equation (31a)which contains o in multipli-
cation form.
7 An intuitive proof for this is the following: dW
v
has variance dv; e
zv
dW
v
has variance
(e
zv
)
2
dv = e
2zv
dv; nally the sum of all independent variables e
zv
dW
v
has a variance equal to
the sum of the variances, which is
_
t
0
e
2zv
dv.
372 Chapter 17
_
t
0
_
s
0
:(uY s) du ds. This double integral is taken over the upper triangle
of the following diagram:
0
s
u t s
t
s
For a given value of sfor a given height s on the ordinatewe inte-
grate all elements :(uY s) du for u from 0 to s; this is
_
s
0
:(uY s) du. Each of
these slices is a function of s for the following two reasons: parameters
enter the function :(uY s)Y and the upper bound of the integral is s. Each
of these slices is then integrated with respect to s from 0 to t to give
_
t
0
_
s
0
:(uY s) du ds. This is our initial double integral.
There are alternatives to this procedure. First, instead of adding
horizontal slices as we did, we could as well add (horizontally) vertical
slices
_
t
s
:(uY s) ds, corresponding, for instance, to the vertical dotted line.
This amounts to calculating the integral
_
t
0
_
t
s
:(uY s) ds du; this, however,
involves an interchange in the order of integration. Second, an alternative
method preserves the initial order of integration: consider the horizontal
dashed line (symmetric to the vertical dotted line and of equal length)
and integrate a slice along this line, with a proviso: change along that line
any point with coordinates (uY s) into its symmetric point (sY u). Our slice
corresponding to the dashed line will thus be
_
t
s
:(sY u) du, equal to
_
t
s
:(uY s) ds; our double integral is equal to
_
t
0
_
t
s
:(sY u) du ds. We can write
the equality:
_
t
0
_
s
0
:(uY s) du ds =
_
t
0
_
t
s
:(sY u) du ds.
For the same reason, and some additional technical reasons related
to its random character, the rst double integral in equation (5) can be
written as
_
t
0
_
t
s
o(sY u) du dW
s
. (The precise justication for this may be
found in the appendix of the Heath, Jarrow, and Morton 1992 paper,
p. 99.) Equation (5) may thus be written in the form of equation (6).
Appendix 17.A.2 Derivation of Ito's formula, with an application to the solution of
two important stochastic dierential equations
1. Ito's formula
We shall give here a heuristic derivation of Ito 's formula. Suppose that X
t
is a stochastic process governed by
dX
t
=
t
dt o
t
dW
t
(1)
where dW
t
is the innitesimal increment of Brownian motion W
t
,
equal to Z
t

dt
, Z
t
@ N(0Y 1). Consider Y
t
= f (X
t
Y t), where f is twice
dierentiable. We want to determine dY
t
. Expanding Y
t
in Taylor series
gives
dY
t
=
f
X
t
dX
t

f
t
dt
1
2
2
f
X
2
t
dX
2
t
2

2
f
X
t
t
dX
t
dt

2
f
t
2
dt
2
_ _
higher order terms in dX
t
Y dtX (2)
Squaring dX
t
, we get
dX
2
t
= o
2
t
dW
2
t
2o
t
t
dW
t
dt
2
t
dt
2
(3)
The last two terms in (3) are of order higher than dt and can be ignored
when dt is suciently small. However, consider the (dW
t
)
2
term, which is
equal to
(dW
t
)
2
= (Z

dt
)
2
= dtZ
2
Y Z@ N(0Y 1)
where Z denotes Z
t
.
The expected value of (dW
t
)
2
is E(dtZ
2
) = dtE(Z
2
) = dt since E(Z
2
) =
1. Its variance is equal to VAR(dtZ
2
) = dt
2
VAR(Z
2
) = 2dt
2
, because
VAR(Z
2
) is equal to 2; indeed, we can write
VAR(Z
2
) = E(Z
4
) E(Z
2
)
_
2
The fourth moment of Z can be calculated as the fourth derivative of the
moment-generating function of Z,
Z
(z) = e
z
2
a2
, at point z = 0; we get
E(Z
4
) = 3, and therefore VAR(Z
2
) = 2.
From the above, we conclude that the variance of (dW
t
)
2
tends toward
zero when dt becomes extremely small; therefore, it ceases to be a random
374 Chapter 17
variable and tends toward the constant E(Z
2
dt) = dt. Alternatively, we
could make use of Chebyshev's inequality. We can write
prob[(dW
t
)
2
dt[ ` b 1
VAR(dW)
2
and, equivalently,
prob[(dW
t
)
2
dt[ ` b 1
2dt
2
which shows that however small a value we may choose for , we can
make dW
2
t
converge toward dt by making dt suciently small. We then
have
dX
2
t
= o
2
t
dW
2
t
= o
2
t
dt
Now dY
t
has a part that tends toward
1
2
2
f
X
2
t
o
2
t
dt when dt becomes very
small. Being of order dt, it cannot be neglected, like the two other higher
order terms

2
f
X
t
t
dX dt and
1
2
2
f
X
2
t
dt
2
. Therefore, the rst-order dierential
of Y
t
is equal to
dY
t
=
f
X
t
dX
t

f
t
dt
1
2
o
2
t
2
f
X
2
t
dt (4)
Plugging dX
t
from (1) into (4), we get
dY
t
=
t
f
X
t
f
t

1
2
o
2
t
2
f
X
2
t
_ _
dt o
t
f
X
t
dW
t
(5)
This is Ito 's lemma. We can verify immediately that from any process
X
t
= X
0

_
t
0
v
dv
_
t
0
o
s
dW
s
(6)
we can deduce dX
t
=
t
dt o
t
dW
t
. Indeed, let us write f (X
t
) = X
t
; then
f
X
t
= 1;

2
f
X
2
t
= 0;
f
t
= 0 and Ito 's formula then leads to
dX
t
=
t
dt o
t
dW
t
(7)
as it should. Thus (6) is the solution, or the diusion, of the stochastic
dierential equation (7).
2. The solution of the two important stochastic dierential equations
2.1 The Doleans equation
Suppose we are given a stochastic dierential equation (SDE) of the fol-
lowing form:
dS
t
S
t
= dt o dW
t
(8)
This is called the Doleans SDE, and it is frequently used in nance. In
order to nd its solution, as is often the case in integration, we need to use
trial and error. We rst try a solution of the form
S
t
= S
0
e
mtsW
t
(9)
and then use Ito 's formula to derive the dierential of S; we next see
whether we can identify m and s such that, if used in (9), the dierential of
S would exactly match (8).
Let us apply Ito 's formula to (9). We have, with X
t
= mt sW
t
and
dX
t
= mdt s dW
t
Y S
t
= S
0
e
X
t
. Note here that
f
t
= 0 because S
t
depends
on time solely through Brownian motion X
t
and does not exhibit addi-
tional dependence upon time. Therefore,
dS
t
= (mS
0
e
X
t
1
2
s
2
S
0
e
X
t
) dt sS
0
e
X
t
dW
t
(10)
or, equivalently, with S
t
= S
0
e
X
t
Y
dS
t
S
t
= (m
1
2
s
2
) dt s dW
t
(11)
We can now identify the coecients of dt and dW
t
in (11) with those of (8)
to obtain the values of m and s. We can write m
1
2
s
2
= and s = o. The
identication of s as o is immediate, and m is given by
o
2
2
. The solution
is then
S
t
= S
0
e

o
2
2
_ _
toW
t
(12)
or, equivalently,
log S
t
= log S
0

o
2
2
_ _
t oW
t
(13)
376 Chapter 17
This implies that
log S
t
aS
0
( ) =
o
2
2
_ _
t oW
t
(14)
Thus, the solution of the SDE (8) implies that the continuously com-
pounded rate of return of S
t
over period [0Y t[ is a Brownian motion with
drift
o
2
2
and variance o
2
.
2.2 The solution of dY
t
Fs(t)Y
t
dW
t
In our last example, we saw that Ito 's lemma led, in an exponential
Wiener process, to subtracting in the exponent the term
1
2
o
2
from . So we
can imagine that a solution to the above equation would be given by
Y
t
= Y
0
exp
_
t
0
o
s
dW
s

1
2
_
t
0
o
2
s
ds
_ _
(15)
Indeed, applying Ito 's lemma to Y
t
will yield the above dierential
equation.
We will now show that a necessary condition for Y
t
to be a martingale
is met. We must have E(Y
t
) = Y
0
X The expectation of Y
t
is
E(Y
t
) = Y
0
exp
1
2
_
t
0
o
2
s
ds
_ _
E exp
_
t
0
o
s
dW
s
_ _ _ _
(16)
The expectation term on the right-hand side of (16) can be written in
the following way: let hs
j
, j = 1, F F F , n be an equally spaced partition
of the interval [0Y t[, and
_
t
0
o(s) dW
s
= lim
ny
n
j=1
o
j
hW
j
, where all hW
j
variables are independent and normal N(0Y hs
j
). We can write
E
_
e
r
t
0
o(s) dW
s
_
= E
_
e
lim
ny
n
j=1
o
j
hW
j
_
= lim
ny
E
_
e
n
j=1
o
j
hW
j
_
Each expectation E(e
o
j
hW
j
) is the moment-generating function of
hW
j
@ N(0Y hs
j
), where o
j
plays the role of the parameter; it is equal to
e
1
2
o
2
j
hs
j
. Thus we have
lim
ny
E
_
e
n
j=1
o
j
hW
j
_
= lim
ny
e
1
2
n
j=1
o
2
j
hs
j
= e
1
2
r
t
0
o
2
(s) ds
and, nally,
E(Y
t
) = Y
0
exp
1
2
_
t
0
o
2
(s) ds
_ _
exp
1
2
_
t
0
o
2
(s) ds
_ _
= Y
0
Thus a necessary condition to be met for Y(t) to be a martingale.
Main formulas
1. A Guide to the Heath-Jarrow-Morton One-Factor
Model: Four Essential Steps8
Hypotheses
.
Evolution of forward rate f (tY T):
t
:(tY T) dt (1)
or
f (tY T) = f (0Y T)
_
t
0
o(sY T) dW
s

_
t
0
:(sY T) ds (2)
.
Particular case: evolution of current, short rate:
r(t) I f (tY t) = f (0Y t)
_
t
0
o(sY t) dW
s

_
t
0
:(sY t) ds (3)
Valuations
.
Cash account:
B(t) = e
r
t
0
r(s) ds
= exp
_ _
t
0
f (0Y u) du
_
t
0
_
t
s
o(sY u) du dW
s

_
t
0
_
t
s
:(sY u) du ds
_
(6)
.
Zero-coupon bond in current value:
8 The numbering of the equations in this summary is that used in the text. We suppose
throughout this ``guide'' that a number of technical conditions on o(tY T) and :(tY T), as
expressed in M. Baxter and A. Rennie (op. cit., p. 143), are met. Also, for the justication
for interchanging of limits in the double integrals involving stochastic dierentials, see the
appendix of Heath, Jarrow, and Morton (op. cit.).
378 Chapter 17
B(tY T) = e
r
T
t
f (tYu) du
= exp
_
_
T
t
f (0Y u) du
_
t
0
_
T
t
o(sY u) du dW
s

_
t
0
_
T
t
:(sY u) du ds
_
(8)
.
Zero-coupon bond in present value:
Z(tY T) = B
1
(t)B(tY T)
= exp
_
_
T
0
f (0Y u) du
_
t
0
_
T
s
o(sY u) du dW
s

_
t
0
_
T
s
:(sY u) du ds
_
(9)
Finding a change in probability measure that makes Z(t, T) a martingale
.
Apply Ito 's lemma to Z(tY T), set v(tY T) =
_
T
t
o(tY u) du:
t

_
T
t
:(tY u) du dt
v
2
2
(tY u) dt
_ _
(12)
.
Set d
~
W
t
equal to
d
~
W
t
= dW
t

1
v(tY T)
_
T
t
:(tY u) du dt
v(tY T)
2
dt IdW
t

t
dt
with the change of measure
t
:
t
=
1
v(tY T)
_
T
t
:(tY u) du
v(tY T)
2
(16)
The fundamental no-arbitrage condition, a restriction on the change of
measure g
t
.
The change of measure
t
must be independent from T; set the dier-
ential of
t
with respect to T equal to zero to obtain
:(tY T) = o(tY T) v(tY T)
t
[ [ (17)
.
In equation (1) replace dW
t
with d
~
W
t
dt and :(tY T) with the right-
hand side of (17).
.
Then
df
t
= o(tY T) d
~
W
t
Thus the drift coecient under the martingale measure must be
_
T
t
o(tY u) duX
2. Application: the case of the Vasicek volatility reduction factor
Drift coecient
:(tY T) = oe
z(Tt)
_
T
t
oe
z(ut)
du = o
2
e
z(Tt)
e
2z(Tt)
z
_ _
(19)
Dierential of forward rate process
df (tY T) = oe
z(Tt)
d
~
W
t

o
2
z
e
z(Tt)
e
2z(Tt)
_ _
dt (20)
Forward process
f (tY T) = f (0Y T) o
_
t
0
e
z(Tv)
d
~
W
v
o
2
2z
2
e
2zT
(e
2zt
1) 2e
zT
(e
zt
1)
_
(21)
Bond's current value
B(tY T) = exp
_
_
T
t
f (0Y u) du
o
2
4z
3
[(1 e
2zt
)(e
2zT
e
2zt
)
4(e
zt
1)(e
zT
e
zt
)[ o
_
T
t
_
t
0
e
z(uv)
d
~
W
v
du
_
(22)
which is shown to be equal to
B(tY T) =
B(0Y T)
B(0Y t)
exp [r(t) f (0Y t)[X(tY T)
o
2
4z
X
2
(tY T)(1 e
2zt
)
_ _
(31)
with
X(tY T) I
1 e
z(Tt)
z
(28)
380 Chapter 17
Questions
17.1. Make sure you can derive the value of the cash account B(t),
opened at time 0 with an initial value of $1, as given by equation (5).
17.2. Make sure you can see why the double integral
_
t
0
_
s
0
:(uY s) du ds can
be written as
_
t
0
_
t
s
:(sY u) du ds in the expression yielding the current
account's value B(t) (equation (6)).
17.3. If you are not already familiar with Ito 's lemma, read section 1 of
appendix 17.A.2 and try to do the intuitive demonstration on your own.
17.4. Using Ito 's lemma, obtain the expression of the dierential dZ
(equation (12)).
17.5. Derive the value of the drift term as expressed in equation (19).
17.6. Show that obtaining the bond's value using Vasicek's volatility
coecient o(tY T) in the Heath-Jarrow-Morton framework implies three
successive denite integrations, and carry them out.
18
THE HEATH-JARROW-MORTON MODEL
AT WORK: APPLICATIONS TO BOND
IMMUNIZATION
All losses are restor'd, and sorrows end.
Sonnets
Chapter outline
18.1 The immunization process 383
18.1.1 The case of a zero-coupon bond 384
18.1.2 Comparison of longer-term hedging portfolios 389
18.2 Immunizing a bond under the Heath-Jarrow-Morton model 391
18.3 Other applications of the Heath-Jarrow-Morton model 401
Overview
In the preceding chapter we described how to obtain, from Wiener pro-
cesses governing the forward rate, a closed form for the price of a zero-
coupon bond at time t, valued as a function of the dierence between the
short rate at time t and the forward rate that had been set by the market
at time 0 for that date. We did so using the Heath-Jarrow-Morton one-
factor model, where only one Brownian motion drove the forward rate.
We will now give an application in the form of immunization. From the
rst part of this book, we remember that replicating a zero-coupon bond
required us to construct another portfolio with a matching duration. Here
we will not be surprised to nd that the hedging portfolio will exhibit a
somewhat dierent structure, and an important question will be, how
dierent is a Heath-Jarrow-Morton portfolio from a classical portfolio,
constructed on the premises that the structure of interest rates receives a
parallel shift? In the case of nonparallel moves of the yield curve, how
dierent are the portfolios, and how dierent are their durations? Impor-
tant conclusions can be derived from such a comparison.
18.1 The immunization process
The sensitivity of the bond to any shock received by the random variable
I(t) = r(t) f (0Y t), at any point in time, can be analyzed as follows: let
I
0
denote an initial value of I at any time t; I
1
denotes a new value that I
can take a very short instant after. Developing B(I
1
) in Taylor series at
point I
0
up to the second order,
B(I
1
) = B(I
0
) B
/
(I
0
)(I
1
I
0
)
1
2
B
//
(I
0
)(I
1
I
0
)
2
terms of higher order in I
1
I
0
(1)
and thus, setting
hB
B
I
B(I
1
)B(I
0
)
B(I
0
)
and I
1
I
0
IhI, a second-order ap-
proximation of the shock received by B is
hB
B
e
B
/
(I
0
)
B(I
0
)
hI
1
2
B
//
(I
0
)
B(I
0
)
(hI)
2
(2)
In the one-factor Heath-Jarrow-Morton model, the expressions
B
/
(I
0
)
B(I
0
)
and
B
//
(I
0
)
B(I
0
)
are obtained by dierentiating equation (31) of our chapter 17.
Because of the exponential nature of (31), these turn out to be simply
B
/
(I
0
)
B(I
0
)
= X(tY T) (3)
and
B
//
(I
0
)
B(I
0
)
= X
2
(tY T) (4)
Thus the relative increase incurred by B is, in second-order approximation,
hB
B
eX(tY T)hI
1
2
X
2
(tY T)(hI)
2
(5)
This relationship enables us to dene easily a hedging policy for any
zero-coupon bond or any coupon-bearing bond or any portfolio of bonds
for that matter. We will rst consider protecting a single zero-coupon
bond and will then turn toward the immunization of a coupon-bearing
bond.
18.1.1 The case of a zero-coupon bond
Suppose we want to protect a bond1 against a shock in r(t), or equiv-
alently, against a shock in I(t). What we want is to constitute a portfolio
1In this section the term ``bond'' will always mean a zero-coupon bond.
384 Chapter 18
replicating the bond to be protected with other bonds (three of them, for
reasons that will appear soon) such that its initial value is zero and its new
value (in the case of a change in I(t)) is also zero. Let V(I
0
) be the value
of that portfolio and n
a
, n
b
, n
c
the number of bonds a, b, c with values
B
a
(I), B
b
(I), and B
c
(I) to be sold (or bought). We have, by denition,
V(I
0
) = B(I
0
) n
a
B
a
(I
0
) n
b
B
b
(I
0
) n
c
B
c
(I
0
) = 0 (6)
When I moves from I
0
to I
1
, V becomes, in quadratic approximation
V(I
1
) eV(I
0
) V
/
(I
0
)hI
1
2
V
//
(I
0
)(hI)
2
(7)
For V(I
1
) to remain equal to V(I
0
) = 0 whatever the change hI, a set of
necessary and sucient conditions is that all the coecients of the poly-
nomial in the right-hand side of (7) are zero:
i) V(I
0
) = 0
ii) V
/
(I
0
) = 0
iii) V
//
(I
0
) = 0
Equation (i) translates as
B(I
0
) n
a
B
a
(I
0
) n
b
B
b
(I
0
) n
c
B
c
(I
0
) = 0 (8)
For simplicity, denote by X the coecient X(tY T) corresponding to the
initial bond and by X
a
, X
b
, X
c
the coecients corresponding to bonds
a, b, and c. Use the fact that B
/
(I
0
) = XB(I
0
); equation (ii) implies, after
simplication,
XB(I
0
) n
a
X
a
B
a
(I
0
) n
b
X
b
B
b
(I
0
) n
c
X
c
B
c
(I
0
) = 0 (9)
Finally, use the relationship B
//
(I
0
) = X
2
B(I
0
); equation (iii) implies
X
2
B(I
0
) n
a
X
2
a
B
a
(I
0
) n
b
X
2
b
B
b
(I
0
) n
c
X
2
c
B
c
(I
0
) = 0 (10)
(The reader can now see why we have to nd three zero-coupons to
complete our portfolio. Developing the portfolio's value V(I) into a
second-order Taylor series entails a second-order polynomial whose
three coecients have to vanishhence a system of three equations,
which in turn implies at least three unknownsthe numbers n
a
, n
b
, and n
c
of additional bonds to make up the portfolio.)
385 The Heath-Jarrow-Morton Model at Work: Applications to Bond Immunization
For sake of convenience, we have rewritten and regrouped equations
(8) to (10) in a slightly simplied form, which enables us to see immedi-
ately the simple matrix structure of the system (in particular, we have not
included the dependencies of the values of the zero-coupons on I):
B
a
n
a
B
b
n
b
B
c
n
c
= B (8a)
X
a
B
a
n
a
X
b
B
b
n
b
X
c
B
c
n
c
= XB (9a)
X
2
a
B
a
n
a
X
2
b
B
b
n
b
X
2
c
B
c
n
c
= X
2
B (10a)
Solving this system yields the number of bonds n
a
, n
b
, and n
c
to be sold
or purchased. Note that the B's are positive and that, for z b 0, the X's
are always positive; this implies that at least one of the additional bonds
will be held short in the portfolio. An example will illustrate the con-
struction of such a portfolio.
Before we proceed with this example, we should reect on the particu-
lar signicance of the z coecient in our model and on the central role
played by the X(tY T) variable. Remember that o(tY T) = oe
z(Tt)
. Thus
z =
1
o
do
dT
: it is the instantaneous rate of decrease of the forward rate's
volatility coecient with respect to the trading horizon T t, longer
horizons having a lower volatility than shorter ones.
When the trading horizon increases by one year, the volatility's relative
increase is, in exact value,
ho(tY T)
o(tY T)
=
e
z(T1t)
e
z(Tt)
e
z(Tt)
=
e
z(T1t)
e
z(Tt)
1 = e
z
1
The volatility increase turns out to be e
z
1. In linear approximation
(developing e
z
1 in a Taylor series limited to its term of order one in
z), it is approximately equal to z; indeed, e
z
1 ez. For instance,
suppose that z = 10%. This implies that the exact relative increase in the
volatility is e
0X1
1 = 0X0952 = 9X52%; thus the volatility coecient
decreases by 9.52% per additional year in the trading period, while it
decreases by 10% in linear approximation.
Let us now turn to the all-important coecient
X(tY T) =
1 e
z(Tt)
z
As we noticed before, the exponential expression for the zero-coupon
bond's valueequation (31) of chapter 17reveals that X(tY T) is simply
386 Chapter 18
minus the sensitivity coecient of B(tY T) with respect to the variable
I(t) = r(t) f (0Y t) and that X
2
(tY T) is the convexity coecient of
B(tY T) with respect to the same variable I(t)X Those properties enable us
to answer easily the question of the units in which X(tY T) is expressed.
Since X is
1
B(tYT)
dB(tYT)
dI
and since dI is expressed in 1/year, it is immediate
that X(tY T) is expressed in years.
From the hypotheses of the model, we know that z = 0 implies that
o(tY T) = o, a constant. Whatever shock interest rates will receive, the
shock will be independent from the time horizon, which means that the
forward rate curve will receive a parallel displacement. Hence, a zero-
coupon bond with maturity (or duration) T t will have a sensitivity
equal to its duration T t. This can be veried by taking the limit of
X(tY T) when z 0; indeed, applying L'Hospital's rule, we have
lim
z0
X(tY T) = lim
z0
1 e
z(Tt)
z
= T t
We now take up again the question we alluded to in this chapter's
introduction: consider the general single-factor Heath-Jarrow-Morton
model where z is not equal to zero (z b 0). Suppose we want to immunize
a zero-coupon bond against shocks in interest rates. How does a hedging
policy based upon the Heath-Jarrow-Morton model dier from a policy
resulting from the classical results about duration? The question makes a
lot of sense, since we may well imagine that a model like the one devel-
oped by Heath, Jarrow, and Morton might yield quite dierent results
compared to those obtained with the classical model.
For easy reference, and in order to have comparable results, the ex-
ample that will serve as our basis for work is the one considered in Jarrow
and Turnbull.2 In this base case, o = 0X01 and z = 0X1; a one-year zero-
coupon is to be hedged with three zero-coupon bonds, with maturities
equal to 0.5, 1.5, and 2 years, respectively. The B, BX, and BX
2
values
are given in table 18.1.
Let us now apply these data to systems (8) through (10) and solve it
in n
a
, n
b
, n
c
. We get, as Jarrow and Turnbull did: n
a
= 0X3093, n
b
=
1X0776, and n
c
= 0X3872. Now there are some interesting questions to
ask: how far is this hedging portfolio from a ``classical'' portfolio that
2R. Jarrow and S. Turnbull, Directive Securities (Cincinnati, Ohio: South-Western, 1996),
p. 494.
would be built on the premise that the whole structure would undergo a
parallel shift? And how dierent is the duration of the hedging portfolio
(which will be referred to henceforth as the HJM portfolio) from the
duration of the classical portfolio (which, of course, is equal to one)?
To answer the rst question, let us recalculate our values X, XB, and
X
2
B when z = 0 (or, equivalently, when X = T t) and solve systems (8)
through (10). We obtain n
a
= 0X325, n
b
= 1X025, and n
c
= 0X351. That
``classical portfolio'' is rather close to the HJM one. Now calculate the
duration of the initial portfolio. Denoting by :
a
, :
b
, and :
c
the shares of
each bond in the portfolio, those shares are
:
a
= n
a
B
a
B
p
= 0X31681
:
b
= n
b
B
b
B
p
= 1X051271
:
c
= n
c
B
c
B
p
= 0X36808
On the other hand, the duration of the portfolio is equal to
D
p
= :
a
D
a
:
b
D
b
:
c
D
c
(that is, it is equal to the average of the durations, weighted by the re-
spective shares of each bond in the portfolio). We thus get
D
p
= (0X31681)0X5 (1X051271)1X5 (0X36808)2 = 0X9992
The duration of the HJM portfolio is practically the same as the
duration of the classical portfolio. This is an interesting property, which is
somewhat at odds with what Jarrow and Turnbull state. In their remark-
able text (p. 495), the authors started by hedging the one-year bond with
Table 18.1
Base case: sF0.01; lF0.1
Maturity Discount rate X
(years) (per year) Price B (years) XB X
2
B
1 0.0458 95.42 0.951626 90.80414 86.41156
0.5 0.0454 97.73 0.487706 47.66348 23.24576
1.5 0.0461 93.085 1.39292 129.66 180.606
2 0.0464 90.72 1.812692 164.4475 298.0927
388 Chapter 18
the rst-order sensitivity coecient only, thereby using two equations in
two unknowns. They had (quite correctly, of course) obtained the port-
folio n
a
= 0X4760 and n
b
= 0X5254. Unfortunately, they were led to
calculate (by some unfortunate typos) the duration of the resulting port-
folio as
0X4760(0X5) 0X5254(1X5) = 1X0261
In this equation, the left-hand side adds two quantities expressed in dif-
ferent units: (0X4760)(0X5) is in (units of bond a) times years; 0X5254(1X5)
is in (units of bond b) times years. These cannot be added, and the
result is not in years, as it should be. The correct calculation should be
the weighted average of the durations, where the weights are the shares
of each zero-coupon in the balancing portfolio. Here those shares are
0X476
97X73
95X42
= 0X4875 and 0X525
93X085
95X42
= 0X5125, and therefore the
duration is
0X4875(0X5) 0X5125(1X5) = 1X0124 years
which yields a duration signicantly closer to onequite a remarkable
result considering that only a linear approximation had been used in the
hedging process. This invites us to examine whether the kind of (true)
proximity we have observed between the duration of an HJM portfolio
and a classical portfolio stands the trial of larger z values, as well as longer
maturities for the bond to be hedged and the bonds of the hedging
portfolio.
We started by considering the same set of zero-coupon bonds as above
and varying the volatility reduction coecient from z = 0X05 to 0X2. The
results in terms of the hedging portfolios, and their durations, are found in
table 18.2.
Note that even if the volatility reduction factor is as high as 20%, the
duration of the hedging portfolio diers from one by less then 1% (3.5 per
thousand means that the order of magnitude of the dierence in durations
between the classical portfolio and the HJM portfolio is of the order of
one daya negligible dierence).
18.1.2 Comparison of longer-term hedging portfolios
Let us now run the test of longer zero-coupon bonds. We have considered
a two-year zero-coupon bond to be hedged with the help of bonds a, b,
and c, whose maturities are twice those considered earlier (1, 3, and 4
years, respectively). The term structure as dened by discount rates and
the corresponding data are shown in table 18.3.
The resulting hedging portfolios and their durations, corresponding to
volatility reduction factor values from z = 0 to z = 2, are represented in
table 18.4.
We can see from these results that for a usual value of the volatility
reduction factor (z = 0X1) the duration of the hedging HJM portfolio is
still less than 1% of a year dierent from the duration of the classical
hedging portfolio (1.993 years as opposed to two years). We have to
double the volatility reduction factor (z = 0X2) to have a dierence of 3%
of a year between those durations.
It is our view that such small dierences stem from a remarkable
property of the Heath-Jarrow-Morton model: it provides the classical
immunization results in the special case of z = 0; and when z goes away
from zero, the immunization process generally starts to dier from the
duration that a classical process would call for, but it does so in a smooth
and regular way, rather than abruptly. Also, it is perfectly logical that
there is an innite number of cases in which, even if the term structure
displacement is not parallel, the durations will match exactly in the clas-
Table 18.3
Data for longer maturities
Maturity Discount Price
2 0.0458 90.84
1 0.0454 95.46
3 0.0461 86.17
4 0.0464 81.44
Table 18.2
Hedging portfolios and their durations. Same bonds; lF0 to 0.2.
HJM portfolios (z b 0)
Number of bonds
in hedging portfolio
Classical
portfolio
z = 0 z = 0X05 z = 0X1 z = 0X15 z = 0X2
n
a
0.325 0.317 0.309 0.301 0.284
n
b
1.025 1.051 1.078 1.105 1.133
n
c
0.351 0.369 0.387 0.407 0.427
Portfolio's duration 1 0.9998 0.9991 0.998 0.9965
390 Chapter 18
sical and the HJM cases. This is, after all, what we would expect from a
model that provides nonparallel shifts through a reduction factor in vol-
atility of the forward rate process.
Our task now is to see whether the same conclusions apply in the
immunization of coupon-bearing bonds.
18.2 Immunizing a bond under the Heath-Jarrow-Morton model
In order to avoid possible confusion between the price of a zero-coupon
bond and the price of a coupon-bearing bond, we will use the following
notation:
P(tY T
n
; I) = value of a bond at time t, with maturity at date T
n
; its
dependency on I = r(t) f (0Y t) is underlined; the letter P stands both
for the price of the bond and for a portfolio of n zero-coupon bonds: the n
payments, or cash ows paid out by the issuer of the bond to its owner;
the rst n 1 are equal to the coupons; the last one, c
n
, is the last coupon
plus the principal.
For short, P(tY T
n
; I) will be abbreviated to P(I) or even P.
c
i
= number of dollars paid by the issuer at time T
i
(i = 1Y F F F Y n).
B(tY T
i
; I) = value at time t of a zero-coupon paying one dollar at time T
i
.
c
i
B(tY T
i
; I) = value at time t of c
i
dollars paid out by the issuer at time T
i
.
We thus have
P(tY T
n
; I) =
n
i=1
c
i
B(tY T
i
; I) (11)
Table 18.4
Hedging portfolios and their durations bond a: 1 year; bond b: 3 years; bond c:
4 years; lF0 to 2.0
Number of bonds
in hedging portfolio
Classical
portfolio
z = 0 z = 0X05 z = 0X1 z = 0X15 z = 0X2
n
a
0.317 0.301 0.285 0.271 0.256
n
b
1.054 1.108 1.165 1.225 1.288
n
c
0.372 0.411 0.351 0.498 0.547
Portfolio's duration
(years) 2 1.998 1.993 1.984 1.971
Under the Heath-Jarrow-Morton one-factor model, the bond's theo-
retical value is thus the portfolio of the n zero-coupons, $1 principal
bonds B(tY T
i
), each of which can be evaluated through formula (31) of
chapter 17.
In order to immunize this coupon-bearing bond, we will proceed in the
same way as before in the case of zero-coupon bonds. To simplify nota-
tion, our variables P(tY T
n
; I) and B(tY T
i
; I) will be written P(I) and
B
i
(I); thus P(I) =

n
i=1
c
i
B
i
(I).
First, develop P(I) in a second-order Taylor series around I
0
. We have
P(I
1
) eP(I
0
) P
/
(I
0
)(I
1
I
0
)
1
2
P
//
(I
0
)(I
1
I
0
)
2
(12)
Setting
hP
P
I
P(I
1
)P(I
0
)
P(I
0
)
and I
1
I
0
IhI, the relative increase in the
bond's value after a shock hI is, with a second-order approximation
hP
P
e
P
/
(I
0
)
P(I
0
)
hI
1
2
P
//
(I
0
)
P(I
0
)
(hI)
2
(13)
Let us now determine the two coecients of this second-order polyno-
mial. First, from (11) we have
P
/
(I
0
)
P(I
0
)
=
n
i=1
c
i
B
/
i
(I
0
)
P(I
0
)
=
n
i=1
c
i
B
i
(I
0
)
P(I
0
)

B
/
i
(I
0
)
B
i
(I
0
)
(14)
Replacing B
/
(I
0
)aB
i
(I
0
) with X(tY T
i
) IX
i
and the shares of each
cash ow in the bond
c
i
B
i
(I
0
)
P(I
0
)
with :
i
(
n
i=1
:
i
= 1), we have
P
/
(I
0
)
P(I
0
)
=
n
i=1
:
i
X
i
IX (15)
Thus the relative change in the coupon-bearing bond is simply minus the
weighted average of the X
i
's, the weights being the share of each of the
cash ows in the bond's value.
Similarly, the second-order term is
P
//
(I
0
)
P(I
0
)
=
n
i=1
c
i
B
//
i
(I
0
)
P(I
0
)
=
n
i=1
c
i
B
i
(I
0
)
P(I
0
)

B
//
i
(I
0
)
B
i
(I
0
)
(16)
We know that each variable B
//
i
(I
0
)aB
i
(I
0
) is simply X
2
(tY T
i
) IX
2
i
.
Thus
392 Chapter 18
P
//
(I
0
)
P(I
0
)
=
n
i=1
:
i
X
2
i
IX
2
(17)
The second-order coecient is the weighted average of the squares of the
X
i
's. The relative change in the bond's value as expressed in (13) can be
written as
hP
P
eXhI
1
2
X
2
P(hI)
2
(18)
and the increase in the bond's value is approximately equal to
hP = XPhI
1
2
X
2
P(hI)
2
(19)
This relationship applies directly to any bond, with corresponding X and
X
2
coecients.
Suppose we now want to protect our coupon-bearing bond whose
value is P(I) against any shock, at time t, received by I. We can think of
building a hedging portfolio by adding positions (positive or negative) in
three bonds, A, B, and C, to our initial bond. Let N
A
, N
B
, and N
C
be the
number of bonds to be determined; the prices of the bonds are P
A
, P
B
,
and P
C
, respectively, and the value of the hedging portfolio is V
h
. Thus, if
our hedging portfolio is to have initial zero value, a rst equation to be
satised by the unknowns N
A
, N
B
, N
C
must be
V
h
(I
0
) = P(I
0
) N
A
P
A
(I
0
) N
B
P
B
(I
0
) N
C
P
C
(I
0
) = 0 (20)
When I undergoes a shock from I
0
to I
1
, the hedging portfolio takes a
new value equal to (in second-order approximation)
V
h
(I
1
) eV
h
(I
0
) V
/
h
(I
0
)hI
1
2
V
//
h
(I
0
)(hI)
2
(21)
For the hedging portfolio to remain xed at V
h
(I
0
) = 0 whatever the
values of hI and (hI)
2
may be, a rst condition is that (20) is met. In
addition, the coecients V
/
h
(I
0
) and V
//
h
(I
0
) in (21) must be equal to zero.
V
/
h
(I
0
) = 0 implies that
V
/
h
(I
0
) = P
/
(I
0
) N
A
P
/
A
(I
0
) N
B
P
/
B
(I
0
) N
C
P
/
C
(I
0
) = 0 (22)
From (15), we know that P
/
(I
0
) = XP(I
0
). As we mentioned earlier,
this relation extends of course to bonds A, B, and C, so we have P
/
A
(I
0
) =
X
A
P
A
(I
0
), and so forth. The above relation can then be written as
V
/
h
(I
0
) = XP(I
0
) N
A
X
A
P
A
(I
0
) N
B
X
B
P
B
(I
0
) N
C
X
C
P
C
(I
0
) = 0
(23)
which implies
XP(I
0
) N
A
X
A
P
A
(I
0
) N
B
X
B
P
B
(I
0
) N
C
X
C
P
C
(I
0
) = 0 (24)
This is the second equation of our system in N
A
, N
B
, N
C
. The third equa-
tion is provided by V
//
h
(I
0
) = 0:
V
//
h
(I
0
) = P
//
(I
0
) N
A
P
//
A
(I
0
) N
B
P
//
B
(I
0
) N
C
P
//
C
(I
0
) = 0 (25)
From (17), we know that P
//
(I
0
) = X
2
P(I
0
), and as for the rst-order
derivative, this relationship extends to any other bond. The preceding
equation may therefore be written as
X
2
P(I
0
) N
A
X
2
A
P
A
(I
0
) N
B
X
2
B
P
B
(I
0
) N
C
X
2
C
P
C
(I
0
) = 0 (26)
The system to be solved corresponds to the following, written in simplied
form:
P
A
N
A
P
B
N
B
P
C
N
C
= P (27)
X
A
P
A
N
A
X
B
P
B
N
B
X
C
P
C
N
C
= XP (28)
X
2
A
P
A
N
A
X
2
B
P
B
N
B
X
2
C
P
C
N
C
= X
2
P (29)
Thus, the linearity of the bonds in their components has provided us
with a system of equations (27) through (29), which is basically the same
as our previous system (8a) through (10a), corresponding to hedging a
zero-coupon. The prices of the zero-coupons (the B's) are replaced by the
prices of the bonds (the P's); the numbers of each individual zero-coupon
(the n's) are replaced by the numbers of each bond; each individual zero-
coupon's X is replaced by X, the weighted average of the X's of each
bond; and nally the square of X is replaced by the weighted average of
all squares X
2
pertaining to the cash ows of each bond.
We rst experimented with the following bond: P, the bond to be
immunized, has a 10-year maturity, and a coupon of 5; the initial term
structure is given by the polynomial
R(0Y T) = (0X005T
2
0X2T 4)a100
(It corresponds to the rst part of our term structure described in chapter
11.)
394 Chapter 18
The bond's price, its X = B
/
aB sensitivity coecient, as well as its
BX and BX
2
values and the initial term structure, are given in table
18.5; tables 18.6, 18.7, and 18.8 provide the same information for the
hedging bonds whose maturities are 7, 8, and 15 years, with coupons of 4,
4.75, and 7, respectively. The common volatility reduction factor ranges
from 0 to 0.2.
The results are presented in table 18.9, in the form of the classical
portfolio (corresponding to z = 0) and the HJM portfolio (z b 0). For
each portfolio, the Fisher-Weil duration is also indicated.
Now watch carefully on the last line of table 18.9, the evolution of the
duration of the HJM portfolio as z increases from z = 0 to z = 0X1.
Contrary to what our intuition would suggest, it does not diverge more
and more from the duration of the classical portfolio: rst, it increases,
and then not only does it decrease, but it falls below the level of the clas-
sical duration. This prompts us to think that, since D
P
is a continuous
function of z, there must be at least one value of z dierent from zero for
which both the classical portfolio and the HJM portfolio have the same
value. Indeed, for a value slightly above z = 0X09 (in fact, z = 0X09357),
the dierence between the durations falls to 10
6
.
This shows that there is not necessarily a dierence between the dura-
tion of an HJM portfolio and that of a classical portfolio, although the
portfolios themselves will be dierent, if ever so slightly. (In the case just
mentioned, the HJM portfolio would be N
A
= 1X444; N
B
= 2X265; and
N
C
= 0X109.) Note also that the nonmonotonicity of the duration is
mirrored in that of the HJM portfolios: between z = 0 and z = 0X1, N
A
rst decreases and then increases; the negative position in bond B rst
decreases and then increases; and the reverse is true for bond C.
We may then well surmise that if, for a given set of bonds, there exists a
value z (H0) for which the classical and the HJM durations are identical,
the converse must be true: for a given value of z, there must be one (and
probably an innitely high) number of coupon sets for which the dura-
tions are the same, if coupons have the same continuity as z. Indeed, for
z = 0X1, for instance, it suces to consider for bond C a coupon of 9.25
for both durations to be separated by less than 4 10
4
. The same rea-
soning would apply if we considered varying the bonds' maturities, keep-
ing z and the coupons as given. Naturally, we could now envision the
T
a
b
l
e
1
8
.
6
B
o
n
d
A
u
s
e
d
f
o
r
i
m
m
u
n
i
z
a
t
i
o
n
T
B
/
a
B
=
X
A
(
T
)
s
+
B
/
a
B
=
s
+
X
A
(
T
)
s
+
B
/
/
a
B
=
s
+
X
2
A
(
T
)
s
+
T
(
y
e
a
r
s
)
R
(
0
Y
T
)
c
+ T
c
+ T
=
c
T
e
R
(
0
Y
T
)
T
s
+ T
I
c
+ T
a
P
A
(
y
e
a
r
s
)
(
y
e
a
r
s
)
(
y
e
a
r
s
2
)
(
y
e
a
r
s
)
1
0
.
0
4
1
9
5
4
3
.
8
3
6
0
.
0
4
1
0
.
9
5
2
0
.
0
3
9
0
.
0
4
1
0
.
0
5
0
2
0
.
0
4
3
8
4
3
.
6
6
5
0
.
0
3
9
1
.
8
1
3
0
.
0
7
2
0
.
0
7
9
0
.
0
9
6
3
0
.
0
4
5
5
5
4
3
.
4
8
9
0
.
0
3
8
2
.
5
9
1
0
.
0
9
7
0
.
1
1
3
0
.
1
3
6
4
0
.
0
4
7
2
4
3
.
3
1
2
0
.
0
3
6
3
.
2
9
7
0
.
1
1
8
0
.
1
4
3
0
.
1
7
3
5
0
.
0
4
8
7
5
4
3
.
1
3
5
0
.
0
3
4
3
.
9
3
5
0
.
1
3
3
0
.
1
6
9
0
.
2
0
4
6
0
.
0
5
0
2
4
2
.
9
6
0
0
.
0
3
2
4
.
5
1
2
0
.
1
4
4
0
.
1
9
1
0
.
2
3
1
7
0
.
0
5
1
5
5
1
0
4
7
2
.
4
9
7
0
.
7
8
5
.
0
3
4
3
.
9
2
9
5
.
4
6
3
0
.
2
5
4
P
A
=
9
2
X
8
9
2
X
A
=
4
X
5
3
1
X
2
A
=
2
1
X
7
5
6
D
A
=
6
X
1
9
8
X
A
P
A
=
4
2
0
X
9
0
1
X
2
A
P
A
=
2
0
2
0
X
9
9
T
a
b
l
e
1
8
.
5
B
o
n
d
t
o
b
e
i
m
m
u
n
i
z
e
d
*
T
B
/
a
B
=
X
P
(
T
)
s
+
B
/
a
B
=
s
+
X
P
(
T
)
s
+
B
/
/
a
B
=
s
+
X
2
P
(
T
)
s
+
T
(
y
e
a
r
s
)
R
(
0
Y
T
)
c
t
c
+ T
=
c
T
e
R
(
0
Y
T
)
T
s
+ T
I
c
+ T
a
P
(
y
e
a
r
s
)
(
y
e
a
r
s
)
(
y
e
a
r
s
2
)
(
y
e
a
r
s
)
1
0
.
0
4
1
9
5
5
4
.
7
9
5
0
.
0
5
0
.
9
5
2
0
.
0
4
8
0
.
0
4
5
0
.
0
5
0
2
0
.
0
4
3
8
5
4
.
5
8
1
0
.
0
4
8
1
.
8
1
3
0
.
0
8
7
0
.
1
5
7
0
.
0
9
6
3
0
.
0
4
5
5
5
5
4
.
3
6
1
0
.
0
4
5
2
.
5
9
2
0
.
1
1
8
0
.
3
0
5
0
.
1
3
6
4
0
.
0
4
7
2
5
4
.
1
4
0
0
.
0
4
3
3
.
2
9
7
0
.
1
4
2
0
.
4
6
9
0
.
1
7
3
5
0
.
0
4
8
7
5
5
3
.
9
1
8
0
.
0
4
1
3
.
9
3
5
0
.
1
6
1
0
.
6
3
2
0
.
2
0
4
6
0
.
0
5
0
2
5
3
.
7
0
0
0
.
0
3
9
4
.
5
1
2
0
.
1
7
4
0
.
7
8
5
0
.
2
3
1
7
0
.
0
5
1
5
5
5
3
.
4
8
5
0
.
0
3
6
5
.
0
3
4
0
.
1
8
3
0
.
9
2
1
0
.
2
5
4
8
0
.
0
5
2
8
5
3
.
2
7
7
0
.
0
3
4
5
.
5
0
7
0
.
1
8
8
1
.
0
3
6
0
.
2
7
3
9
0
.
0
5
3
9
5
5
3
.
0
7
7
0
.
0
3
2
5
.
9
3
4
0
.
1
9
0
1
.
1
3
0
0
.
2
8
9
1
0
0
.
0
5
5
1
0
5
6
0
.
5
8
0
0
.
6
3
2
6
.
3
1
1
3
.
9
9
3
2
5
.
2
3
8
6
.
3
1
6
P
=
9
5
X
9
1
4
X
=
5
X
2
8
3
X
2
=
3
0
X
7
1
9
D
=
8
X
0
2
3
X
P
=
5
0
6
X
7
1
7
X
2
P
=
2
9
4
6
X
3
5
+
E
x
c
e
p
t
f
o
r
R
(
0
Y
T
)
,
a
l
l
n
u
m
b
e
r
s
i
n
t
h
i
s
t
a
b
l
e
a
n
d
i
n
t
h
e
f
o
l
l
o
w
i
n
g
o
n
e
s
h
a
v
e
b
e
e
n
r
o
u
n
d
e
d
t
o
t
h
e
t
h
i
r
d
d
e
c
i
m
a
l
.
396
T
a
b
l
e
1
8
.
7
B
o
n
d
B
u
s
e
d
f
o
r
i
m
m
u
n
i
z
a
t
i
o
n
T
B
/
a
B
=
X
B
(
T
)
s
+
B
/
a
B
=
s
+
X
B
(
T
)
s
+
B
/
/
a
B
=
s
X
2
B
(
T
)
s
+
T
(
y
e
a
r
s
)
R
(
0
Y
T
)
c
T
c
+ t
=
c
T
e
R
(
0
Y
T
)
T
s
T
=
c
T
a
P
B
(
y
e
a
r
s
)
(
y
e
a
r
s
)
(
y
e
a
r
s
2
)
(
y
e
a
r
s
)
1
0
.
0
4
1
9
5
4
.
7
5
4
.
5
5
5
0
.
0
4
7
0
.
9
5
2
0
.
0
4
5
0
.
0
4
3
0
.
0
4
7
2
0
.
0
4
3
8
4
.
7
5
4
.
3
5
1
0
.
0
4
5
1
.
8
1
3
0
.
0
8
2
0
.
1
4
9
0
.
0
9
0
3
0
.
0
4
5
5
5
4
.
7
5
4
.
1
4
3
0
.
0
4
3
2
.
5
9
2
0
.
1
1
2
0
.
2
8
9
0
.
1
2
9
4
0
.
0
4
7
2
4
.
7
5
3
.
9
3
3
0
.
0
4
1
3
.
2
9
7
0
.
1
3
5
0
.
4
4
4
0
.
1
6
4
5
0
.
0
4
8
7
5
4
.
7
5
3
.
7
2
3
0
.
0
3
9
3
.
9
3
5
0
.
1
5
2
0
.
5
9
9
0
.
1
9
3
6
0
.
0
5
0
2
4
.
7
5
3
.
5
1
5
0
.
0
3
7
4
.
5
1
2
0
.
1
6
5
0
.
7
4
4
0
.
2
1
9
7
0
.
0
5
1
5
5
4
.
7
5
3
.
3
1
1
0
.
0
3
4
5
.
0
3
4
0
.
1
7
3
0
.
8
7
2
0
.
2
4
1
8
0
.
0
5
2
8
1
0
4
.
7
5
6
8
.
6
6
1
0
.
7
1
4
5
.
5
0
7
3
.
9
3
1
2
1
.
6
4
5
5
.
7
1
0
P
B
=
9
6
X
1
9
2
X
B
=
4
X
7
9
5
X
2
B
=
2
4
X
7
8
6
D
B
=
6
X
7
9
5
X
B
P
B
=
4
6
1
X
1
9
5
X
2
B
P
B
=
2
3
8
4
X
1
5
397
T
a
b
l
e
1
8
.
8
B
o
n
d
C
u
s
e
d
f
o
r
i
m
m
u
n
i
z
a
t
i
o
n
T
B
/
a
B
=
X
C
(
T
)
s
+
B
/
a
B
=
s
+
X
C
(
T
)
s
+
B
/
/
a
B
=
s
X
2
C
(
T
)
s
+
T
(
y
e
a
r
s
)
R
(
0
Y
T
)
c
T
c
+ T
=
c
T
e
R
(
0
Y
T
)
T
s
+ T
=
c
+ T
a
P
C
(
y
e
a
r
s
)
(
y
e
a
r
s
)
(
y
e
a
r
s
2
)
(
y
e
a
r
s
)
1
0
.
0
4
1
9
5
7
6
.
7
1
2
0
.
0
6
0
0
.
9
5
2
0
.
0
5
7
0
.
0
5
4
0
.
0
6
0
2
0
.
0
4
3
8
7
6
.
4
1
3
0
.
0
5
7
1
.
8
1
3
0
.
1
0
4
0
.
1
8
9
0
.
1
1
5
3
0
.
0
4
5
5
5
7
6
.
1
0
6
0
.
0
5
5
2
.
5
9
2
0
.
1
4
2
0
.
3
6
8
0
.
1
6
4
4
0
.
0
4
7
2
7
5
.
7
9
6
0
.
0
5
2
3
.
2
9
7
0
.
1
7
1
0
.
5
6
5
0
.
2
0
8
5
0
.
0
4
8
7
5
7
5
.
4
8
6
0
.
0
4
9
3
.
9
3
5
0
.
1
9
3
0
.
7
6
1
0
.
2
4
6
6
0
.
0
5
0
2
7
5
.
1
8
0
0
.
0
4
6
4
.
5
1
2
0
.
2
0
9
0
.
9
4
5
0
.
2
7
9
7
0
.
0
5
1
5
5
7
4
.
8
8
0
0
.
0
4
4
5
.
0
3
4
0
.
2
2
0
1
.
1
0
8
0
.
3
0
6
8
0
.
0
5
2
8
7
4
.
5
8
8
0
.
0
4
1
5
.
5
0
7
0
.
2
2
6
1
.
2
4
7
0
.
3
2
9
9
0
.
0
5
3
9
5
7
4
.
3
0
8
0
.
0
3
9
5
.
9
3
4
0
.
2
2
9
1
.
3
6
0
0
.
3
4
7
1
0
0
.
0
5
5
7
4
.
0
3
9
0
.
0
3
6
6
.
3
2
1
0
.
2
2
9
1
.
4
4
6
0
.
3
6
2
1
1
0
.
0
5
5
9
5
7
3
.
7
8
3
0
.
0
3
4
6
.
6
7
1
0
.
2
2
6
1
.
5
0
9
0
.
3
7
3
1
2
0
.
0
5
6
8
7
3
.
5
4
1
0
.
0
3
2
6
.
9
8
6
0
.
2
2
1
1
.
5
5
0
0
.
3
8
0
1
3
0
.
0
5
7
5
5
7
3
.
3
1
3
0
.
0
3
0
7
.
2
7
5
0
.
2
1
6
1
.
5
7
1
0
.
3
8
6
1
4
0
.
0
5
8
2
7
3
.
0
9
9
0
.
0
2
8
7
.
5
3
4
0
.
2
0
9
1
.
5
7
7
0
.
3
8
9
1
5
0
.
0
5
8
7
5
1
0
7
4
4
.
3
2
6
0
.
4
0
7
.
7
6
9
3
.
0
8
7
2
3
.
9
7
8
5
.
9
6
0
P
C
=
1
1
1
X
5
6
8
X
C
=
5
X
7
4
2
X
2
C
=
3
8
X
2
2
8
D
C
=
9
X
9
0
4
X
C
B
=
6
4
0
X
6
0
5
X
2
C
B
=
4
2
6
5
X
0
8
398
Table 18.9
Immunization results
Number of
bonds
Classical
portfolio
z = 0 z = 0X05 z = 0X1 z = 0X15 z = 0X2
N
a
1.487 1.428 1.453 1.601 1.896
N
b
2.314 2.245 2.277 2.469 2.878
N
c
0.102 0.113 0.107 0.064 0.043
Duration
(years) 8.023 8.027 8.020 7.952 7.733
0 0.05 0.1 0.15 0.2 0.25
2
1
0
1
2
3
N
A
N
C
N
B
4
Figure 18.1
Classical (lF0) and Heath-Jarrow-Morton (lI0) hedging portfolios. The portfolio on the
dotted line corresponds to lF0.09357, and, although somewhat dierent, the classical port-
folio (lF0) has exactly the same duration.
whole spectrum of the z, coupons, and maturities. From our experience
with numerical examples, we can conclude that
.
in general there will be a dierence in duration and composition between
the HJM portfolio and the classical portfolio;
.
these dierences in composition and in duration are not monotonic
functions of z, and convergence may be the case even if z increases;
.
there is an innite number of combinations of z, coupons, and matur-
ities of the immunizing bonds for which there will be no dierence be-
tween durations of the classical portfolio and the HJM portfolio.
We do not consider the fact that the durations may be the same as a
sign of weakness in the HJM model. On the contrary, this kind of simi-
larity in one aspect of the immunizing portfolios is exactly what we may
expect from a model that smoothly generalizes the classical one, allowing
for a reduction in the volatility factor of the forward rate with respect to
0 0.05 0.1 0.15 0.2 0.25
9
8
7
6
5
4
3
2
1
0
D
u
r
a
t
i
o
n

(
y
e
a
r
s
)
Figure 18.2
Duration of classical (lF0) and Heath-Jarrow-Morton (lF0) hedging portfolios as a func-
tion of l. The HJM portfolio indicated by the dashed line (for lF0X09357) is slightly dierent
from the classical portfolio but has the same duration.
400 Chapter 18
maturity (at any xed time) or, equivalently, allowing an increase in that
volatility factor with respect to time (for any xed maturity). Indeed, one
might conjecture that in some circumstances classical immunization will
produce the same benecial eects as the HJM immunization, and we
have shown some of them precisely.
18.3 Other applications of the Heath-Jarrow-Morton model
One of the great advantages of the Heath-Jarrow-Morton model is that it
provides a unied framework under which it is possible to evaluate inter-
est rate derivatives. Indeed, since the evolution of bond prices is described
under their martingale probabilities, it suces to consider, for the valua-
tion of those derivatives, the expected value of the derivative's payo under
those martingale probabilities. The following derivatives can be priced in
this way: futures on Treasury bills and Treasury bonds; options on Trea-
sury bills, on Treasury bonds, on Treasury bill futures, and on Treasury
bond futures; caps, oor, collars, and swaptions. The reader is referred to
chapters 16 and 17 in Jarrow and Turnbull (1995), which give all the
formulas for these derivatives and furthermore provide software to per-
form the necessary calculations.
Main formulas
B
/
(I
0
)
B(I
0
)
= X(tY T) (3)
B
//
(I
0
)
B(I
0
)
= X
2
(tY T) (4)
with
I(t) Ir(t) f (0Y t)
X(tY T) I
1 e
z(Tt)
z
hB
B
eX(tY T)hI
1
2
X
2
(tY T)(hI)
2
(5)
Using the following notation:
P(tY T
n
; I) = value of a bond at time t, with maturity at date T
n
; its
dependency on I = r(t) f (0Y t) is underlined; the letter P stands both for
the price of the bond, and for a portfolio of n zero-coupon bonds: the n
payments, or cash ows paid out by the issuer of the bond to its owner;
the rst n 1 are equal to the coupons; the last one, c
n
, is the last coupon
plus the principal. For short, P(tY T
n
; I) will be abbreviated to P(I) or
even P.
c
i
= number of dollars paid by the issuer at time T
i
(i = 1Y F F F Y n).
B(tY T
i
; I) = value at time t of a zero-coupon paying one dollar at time T
i
.
c
i
B(tY T
i
; I) = value at time t of c
i
dollars paid out by the issuer at time T
i
.
P(tY T
n
; I) =
n
i=1
c
i
B(tY T
i
; I) (11)
hP
P
e
P
/
(I
0
)
P(I
0
)
hI
1
2
P
//
(I
0
)
P(I
0
)
(hI)
2
(13)
With :
i
=
c
i
B
i
(I
0
)
P(I
0
)
,
n
i=1
:
i
= 1:
P
/
(I
0
)
P(I
0
)
=
n
i=1
:
i
X
i
IX (15)
hP
P
eXhI
1
2
X
2
(hI)
2
(18)
A hedging portfolio comprising N
A
, N
B
, and N
C
bonds of types A, B,
and C must solve the system:
P
A
N
A
P
B
N
B
P
C
N
C
= P (27)
X
A
P
A
N
A
X
B
P
B
N
B
X
C
P
C
N
C
= XP (28)
X
2
A
P
A
N
A
X
2
B
P
B
N
B
X
2
C
P
C
N
C
= X
2
P (29)
Questions
18.1. Using the same bonds and discount rates as in Jarrow and Turnbull
(1995, p. 494; see table 18.1 in this chapter), determine the hedging Heath-
402 Chapter 18
Jarrow-Morton portfolios in the cases z = 0X075, z = 0X125, z = 0X175,
z = 0X225, and in the classical case (z = 0). Determine also the duration
of those portfolios.
18.2. Suppose you want to immunize a bond with maturity 10 years, par
value 100, bearing a 5% coupon. You want to replicate this bond with
three bonds (a, b, and c) with the following characteristics:
Maturity Coupon rate Par value
Bond a 7 years 4.25% 100
Bond b 8 years 4.5% 100
Bond c 15 years 9% 100
Suppose also that the initial term structure is given by
R(0Y T) = (0X005T
2
0X2T 4)a100
for T = 1Y F F F Y 15. (This is the term structure we used in section 11.3 of
chapter 11.)
What would be the replicating classical portfolio and the Heath-
Jarrow-Morton portfolios for z = 0X1, z = 0X15, z = 0X2, z = 0X25, and
z = 0X3? Determine the durations of the respective portfolios. What do
you observe? Do your results, for instance, suggest that there might exist a
nonzero value for z and therefore a Heath-Jarrow-Morton portfolio the
duration of which would be the same as the classical portfolio's duration?
BY WAY OF CONCLUSION: SOME
FURTHER STEPS
Queen. Having thus far proceeded
. . . is't not meet
That I did amplify my judgment in
Other conclusions?
Cymbeline
In the past thirty years, research in nance has essentially been founded
on Gaussian statistical processes, which have their roots in Bachelier's
modeling of Brownian motion at the beginning of the twentieth century.1
It is now recognized more and more by academics and practitioners
alike that Wiener geometric processesand generally the lognormality
hypotheseshave been used so extensively mainly for two reasons. First,
they rest on one of the most powerful results of statistics obtained in these
past two centuries: the central limit theorem. Second, they are extremely
convenient from a mathematical point of view. Unfortunately, these pro-
cesses do not describe well the uncertainty of nancial markets, and they
are poorly suited to representing the large swings that have come to be
systematically observed in the value of either individual assets or whole
market indices.
Today theorists and practitioners have concluded that more sophisti-
cated tools are needed to deal with what is observed in the markets: dis-
continuitiesor jumpsin the assets' value, more frequent observations
of very small (either positive or negative) returns than assumed by the
normal distribution, and more frequent steep downfalls or upward moves
than warranted by the norm. However, for nearly four decades, research
spawned from earlier work by Beno t Mandelbrot2 has called for the
use of statistical distributions that would better conform to the observed
1L. Bachelier, ``Theorie de la speculation,'' Annales scientiques de l'Ecole normale super-
ieure, 17 (1900): 2186.
2See, in particular, B. Mandelbrot, ``The Variation of Certain Speculative Prices,'' Journal of
Business, 36 (1963): 394419; B. Mandelbrot, Fractals and Scaling in Finance: Discontinuity,
Concentration, Risk (New York: Springer, 1997); E. Fama, ``Mandelbrot and the Stable
Paretian Hypothesis,'' in The Random Character of Stock Market Prices, ed. P. Cootner
(Cambridge, Mass.: MIT Press, 1964); E. Fama, ``The Behavior of Stock Market Prices,''
Journal of Business, 38 (1965); E. Fama, ``Portfolio Analysis in a Stable Paretian Market,''
Management Science, 11 (1965); and P. A. Samuelson, `Ècient Portfolio Selection for
Pareto-Levy Investments,'' Journal of Financial and Quantitative Analysis, June (1967).
behavior of markets. These are so-called fractal distributions. In addition
to meeting the above-mentioned properties, they share the property of
being time-homothetic in the sense that, over dierent time scales, their
statistical characteristics remain the same. They are, however, very much
at odds with our intuition, not to mention our wishes: in particular, they
generate continuous functions that are nowhere dierentiable.3 This is
also true of a Brownian motion generated by a normal distribution. But
with regard to the dierence of the normal distribution, their density
functions are not known; only their characteristic functions are, which
makes them much more dicult to deal with.
On the other hand, as we saw earlier, the distribution of a sum, or an
average of random variables, converges toward a normal distribution (the
central limit theorem would apply) only if, among other conditions, the
individual probability distributions are independent. Independence was,
and still is, very dicult to justify. A lot of research has investigated
autoregressive processes, giving rise to frequency distributions that also
have the necessary characteristics called for (fat tails, slim center). These
are the ARCH (for autoregressive conditional heteroscedasticity) pro-
cesses and their multiple descendants. A good synthesis of these strands of
thought, as well as a strong call for uniting them, can be found in Edgar
Peters.4 Regarding the current extreme volatility of nancial markets, the
reader should refer to the very convincing book by Robert Shiller.5 As an
example of recent research in fractal theory, we refer the reader to the
work of Jean-Philippe Bouchaud and Marc Potters6, who use a truncated
3The prole of a mountain, or of a mountain chain, constitutes a very good example of a
fractal and how fractal objects can have a deceiving appearance. From far away, we could
think of mountain chains as sets of triangles and estimate the diculty to climb them by the
steepness of their ``slope'' as seen from such distance. Unfortunately for amateur alpinists,
there is no such thing as the slope of a fractal. After six unsuccessful attempts to climb the
Matterhorn from its Italian side, which looked `èasier,'' the British draftsman turned
climber Edward Whymper decided to circle the whole mountain, carefully drawing its fractal
structure from all edges. He noticed two things: the fractal structure was basically the same
on both the Swiss and the Italian sides; but the strata of the whole mountain had, for some
geological reason, been tilted in such a way that the short, nearly vertical steps were more
pronounced on the Italian side (sometimes leading to overhangs), although the average slope
of some Italian edges looked, and was, lower than on the Swiss side. He then decided to try
the Swiss route and was the rst to reach the summit in 1865.
4E. Peters, Fractal Market Analysis (New York: Wiley, 1994).
5R. Shiller, Irrational Exuberance (Princeton, N.J.: Princeton University Press, 2000).
6J.-Ph. Bouchaud and M. Potters, Theorie des risques nanciers (Paris: Alea Saclay, 1997),
pp. 134135.
406 By Way of Conclusion: Some Further Steps
Levy law. This law was itself proposed in the form of a general charac-
teristic function by I. Koponen.7
Why was the fractal approach not more in use? The reasons probably
go back to the very origins of our attempts to comprehend the laws of
nature. Quite understandably, we started trying to solve complex prob-
lems by making simple assumptionsbut these were unfortunately some-
what awed. In his path-breaking essay on growth theory, Robert Solow
wrote: `Àll theory makes assumptions which are not quite true. This is
what makes it theory.'' He added: ``The art of successful theorizing is to
make the inevitable simplifying assumption in such a way that the nal
results are not very sensitive. . . . When the results of a theory seem to ow
specically from a special crucial assumption, then if the assumption is
dubious, the results are suspect.''8
Here we are encountering a case in which the results are so sensitive
to assumptions that a revision of the hypotheses is needed. Since such a
revision has been felt necessary for many years, a legitimate question to
ask is why it was not already undertaken and generalized long ago. Pos-
sibly a number of reasons are at play, and we will try to delineate some
of them. Quite a profound one, in our opinion, has been suggested by
E. Peters.9 His reasoning makes us go back to the very origin of scientic
thinking in Western civilization. Through their philosophers, mathema-
ticians, and geometers, ancient Greeks formed the idea of a perfect uni-
verse, corresponding to simple (and we may add, often harmonious in an
ultimate sense) constructs. In doing so, they were also led by practical
considerations: if the area of a eld had to be measured, the easiest route
was to assimilate its shape to a simple geometric gure, or to a set of
those, whose areas could be easily calculated. They knew that nature did
not conform to their models, but as E. Peters put it very well, they turned
the conclusions around: ``the problem was not with the geometry, but
with the world itself.''10
A good illustration of the problems we inherited from the ancient
Greeks is the way mathematics has been taught since the Renaissance:
most functions (whether dependent on one or several variables) are
smooth in the sense that they are dierentiable. One property of those
7I. Koponen, Physical Review, E, 52 (1995): 1197.
8R. M. Solow, `À Contribution to the Theory of Economic Growth,'' The Quarterly Journal
of Economics, 70 (1956): 439469.
9Peters, op. cit.
10 Peters, op. cit., p. 3
functions is that if you constantly magnify a given portion of their corre-
sponding curves, you ultimately obtain a straight line (or, equivalently, a
plane). Unfortunately, nature does not possess that property: a natural
object retains its complexity at every scale, and physicists tell us that its
complexity can even increase and not necessarily diminish as science has
suggestedand perhaps hopedsince Pericles's century.
In nance, and especially in the subject area of this book, simplicity
was certainly an important factor for the central limit theorem to be
used so extensively. Also, the theorem served as a powerful vehicle for the
so-called ecient market hypothesis, according to which information was
continuously reected by agents in price formation, with new information
reaching us in a random fashion. In fact, it seems that one should recog-
nize three things: information is not spread out uniformly; even if infor-
mation is the same for all actors in the market, it is not interpreted by
all in the same way; and, nally, even if this information conveys the
same message to everybody, it will have dierent consequences on agents
according to their horizon. (For instance, facing a given set of news, short-
term traders may behave quite dierently than pension-fund managers.)
These dierences in information, analysis, and horizon, far from entailing
disequilibrium, on the contrary work together toward establishing equi-
librium prices. In this sense we have a reconciliation between determinism
and randomness. The two can coexist, leading to a market equilibrium
that is disrupted precisely when everybody thinks in the same way, thus
giving rise to abrupt changes and even discontinuities that can be of major
proportions, until diversity restores some degree of stability.
We take this opportunity to indulge in a conjecture. A primary goal of
social scientists is to describe and explain the progress of societies. At an
even more general level, a perennial question has been to discover the
causes of the rise and eventual decline of civilizations. To the best of our
knowledge, the rst thinker who asked the question and came up with
an answer was the great Arab historian Ibn Khaldun (1376)see his
masterly work.11
11Ibn Khaldun's work was translated into English for the rst time by Franz Rosenthal
under the title The Muqaddimah: An Introduction to History (New York: Bollingen, 1958). A
revised edition has been published by Princeton University Press (2nd edition, 3 volumes,
1980). For a discussion of Ibn Khaldun's contribution to the questions we alluded to, we take
the liberty of refering the reader to our work (O. de La Grandville, Theorie de la croissance
economique (Paris, Masson, 1977).)
It is quite possible that some day historians, sociologists, political sci-
entists, andyeseven economists will communicate with each other.
They may then discover that, like natural phenomena and human behav-
ior, the evolution of civilizations conforms to fractal patterns. (Would Ibn
Khaldun be surprised? Maybe not.) We could then witness the beginning
of a reconciliation between long-term determinism due to the will of the
human being, and short-term randomness. Will that be good news or bad
news? Probably good news, because we will be better prepared to face
what we have come to call `àccidents of history.''
ANSWERS TO QUESTIONS
Don Adrian de Armado. I do say you are quick in answers.
Chapter 1
1.1. Applying equation (1) in the text we get 9.28% per year.
1.2. From (1) or (2), you can determine B
t
= B
T
a[1 i(T t)[ =
$97X087.
1.3. From (4), i = 5X396% per year.
1.4. From (3), we can determine B
t
as B
T
(1 i)
(Tt)
= $82X989.
1.5. Applying (7) with t = 2 years and T = 5 years, we get: f (0Y 2Y 5) =
6X67% per year. The dollar ve-year forward rate, 1.06, is the geometrical
average between the dollar two-year spot rate (1.05) and the dollar for-
ward rate (1.0667), with weights
2
5
and
3
5
, respectively. We have indeed
1X06 = (1X05)
2a5
(1X0667)
3a5
.
1.6. Transforming (6) or (7), we have
i(0Y t) =
[1 i(0Y T)[
Tat
1 f (0Y tY T)
(Tt)at
1 = 6X75%
Chapter 2
2.1. a. Any bond, whatever its par value, coupon rate, or (nonzero)
maturity, will increase in value whenever interest rates decrease, because,
fundamentally, it pays xed amounts in a world in which interest rates
arein our hypothesisdecreasing. We also know that B(i)aB
T
is the
weighted average of
caB
T
i
and 1. So when the interest rate is equal to the
coupon rate, the bond's value is at par, that is, $1000. This result is inde-
pendent of the bond's coupon rate and maturity.
b. The bond's value will be farthest from its par value when i = 6%
because of the convexity of the curve B = B(i). (A bond's value is the sum
of convex functionssee the open-form equation (11) or gure 2.3).
c. The bond's value in each case i = 12%, i = 9%, and i = 6% (those for
i = 12% and i = 6% are calculated with the closed-form (15)) are $830.5,
$1000, and $1220.8, respectively, thus conrming what we could infer in
(a) and (b). They would then be directly proportional to the coupon.
2.2. The price of bond B will be lower than the price of bond A by less
than 20% because the bonds' prices are given by
B =
c
i
[1 (1 i)
T
[ B
T
(1 i)
T
which is not a linear function of coupon c but an ane function of c with
a positive, additive term B
T
(1 i)
T
, which is independent of c. Cases in
which the price of B is inferior to the price of A by exactly 20% are the
cases in which either B
T
= 0 or T y. (In the latter case, the bond
would be called a consol.)
2.3. In this riskless world, the equilibrium expected return is i = 6% per
year. The current price of the asset is, either from equation (17) (setting
= 0) or from gure 2.4, p
0
= E(q p
1
)a(1 i) = $105,000a1X06 =
$99,056.
2.4. From (17) or gure 2.4, p
0
would be worth an amount equal to:
p
0
= E(q p
1
)a(1 i ) = $105,000a1X16 = 90,517.
2.5. From (17) or gure 2.4, the consequences are the following:
a. E(R) increases
b. E(R) increases
c. E(R) decreases
Chapter 3
3.1. The reason for this can best be seen by transforming the denitional
equation of i
+
, (1), into (2). Then, the left-hand side of (2) is a polynomial
in i
+
of order T, whose complete expression is given by a Newton bino-
mial expansion of order T. Its right-hand side is the sum of T poly-
nomials of order T 1, T 2, F F F , 0 and therefore a polynomial of order
T 1. Consequently, equation (2) amounts to a polynomial of order T.
This should be solved in order to obtain i
+
.
3.2. If i
1
= i
0
, B
1
= B
0
and from (5) r
H
= i
0
for any horizon H.
3.3. When the bond was bought, its price was B(0X08) = $1112X6. One
day later, this value increases to the par value (B(0X09) = $1000). The
horizon rates of return are given by
412 Answers to Questions
R
H
=
B(i)
B
0
_ _
1aH
(1 i) 1
and, respectively, by
a. R
H=1
=
1000
1112X6
_ _
(1X09) 1 = 2X0%
b. R
H=4
=
1000
1112X6
_ _
1a4
(1X09) 1 = 6X1%
c. R
H=10
=
1000
1112X6
_ _
1a10
(1X09) 1 = 7X8%
3.4. From equation (5), lim
Hy
r
H
= 1 i
1
1 = i
1
. The innite hori-
zon rate of return turns out to be the new interest rate i
1
. This has an
intuitive explanation: whatever a nite maturity T for a bond, any change
in the value of the bond entailed by a change of interest from i
0
to i
1
is of
no consequence to the par value of the bond, which will be reinvested for
an innite period at rate i
1
. Consequently, the innite horizon rate of
return is simply i
1
.
3.5. Calculate B
1
, using a closed-form formula such as (13) in chapter 2,
and then r
H
using equation (5) of chapter 3.
3.6. Such a horizon may indeed exist if the capital loss due to an increase
in the rates of interests is exactly compensated by gains on the reinvest-
ment of coupons and, symmetrically, if the capital gain due to a decrease
in the interest rates is exactly compensated by losses on the reinvestment
of coupons. This is what happens with a horizon equal to 7.7 years.
Intuitively, we can understand that compared to the maturity of the bond,
this horizon cannot be too longotherwise the eect of the reinvestment
of coupons would dominate the change of price eectand it cannot be
too shortotherwise the price eect would be stronger than the reinvest-
ment eect.
Chapter 4
4.1. In the left-hand side of equation (6), the units of B cancel out; we are
left with units of 1ai, which are 1a(1ayear) = years. In the right-hand side
of (6), the units of c
t
cancel out with those of B; 1 i is without units
(because in it i has been multiplied by 1 year). Therefore, the right-hand
side of (6) is also in years, as it should be.
4.2. The units of modied duration (see the rst equation of section 4.4)
are years because 1 i is unitless.
4.3. In the equation of the same section giving the value of
1
B
dB
di
, multiply
by (1 i) to get D. Then, in the resulting equation, factor out (1 i)
T
in both the numerator and the denominator of the right-hand side.
4.4. Durations are best calculated using the closed-form equation (11).
a. For i = 9%, D = 6X995 years (the case of this chapter).
b. For i = 8%, D = 7X098 years.
From the negatively sloped relationship between interest rates and
duration demonstrated in section 4.5, one could infer that duration would
be somewhat larger when i gets smaller.
4.5. The durations of bonds A and B are, respectively,
D
A
= 7X888 years
D
B
= 5X268 years
One can infer from these gures that the relative decrease in value of
bond A will be higher than that of bond B. Indeed, we rst get the fol-
lowing relative changes in exact value:
Value of bond for
i
0
= 12% i
1
= 14%
Relative change in bond's
value, in exact value
Bond A $863.8 $754.3 hB
A
aB
A
= 12X7%
Bond B $908.7 $828.5 hB
B
aB
B
= 8X8%
In linear approximation, those relative changes in value are
dB
A
B
A
=
1
1 i
0
D
A
di =
1
1X12
(7X888)(2%) = 14X1%
and
dB
B
B
B
=
1
1 i
1
D
B
di =
1
1X12
(5X268)(2%) = 9X4%
From the convexity of B(i), one can infer that the relative changes fol-
lowing an increase in interest rates are less pronounced in (absolute) exact
value than in linear approximation, which is conrmed by our results.
4.6. The durations are the following: D
A
= 8X36 years and D
B
= 8X94
years. Since bond A's coupon rate is higher than bond B's, A has a
smaller duration than B's duration and will react less than B to a change
in interest rates. This is conrmed in the following results (in exact value):
Value of bond for
i = 12% i = 14%
B
A
$1000 $867.5 hB
A
aB
A
= 13X3%
B
B
$701.2 $602.6 hB
B
aB
B
= 14X1%
4.7. The durations of bonds A and B are the following: D
A
= 8X602 years
and D
B
= 4X981. Bond A's duration is considerably higher than bond B's,
for the following two reasons: its maturity is longer (without reaching the
levels such that the corresponding duration decreases) and the coupon
rate is smaller. Therefore A will be more sensitive to an increase in interest
rates, as conrmed by the results in the following table:
Value of bond for
i = 12% i = 14%
B
A
(i) $850.6 $735.1 hB
A
aB
A
= 13X6%
B
B
(i) $1091.3 $1000 hB
B
aB
B
= 8X4%
4.8. Compared to the situation described in question 4.5., increasing the
maturity of bond B from 7 to 11 years and reducing its coupon rate from
10% to 5% has a powerful increasing eect on the bond's duration in such
a way that the bonds now have approximately the same duration:
D
A
= 7X888 years and D
B
= 7X898 years. One can infer that their sensi-
tivity to interest rate changes will be approximately the same.
Value of bond for
i = 12% i = 14%
B
A
(i) $863.8 $754.3 hB
A
aB
A
= 12X7%
B
B
(i) $584.4 $509.3 hB
B
aB
B
= 12X9%
Although both bonds are of a very dierent nature, their level of inter-
est rate riskiness is indeed about the same.
4.9. We obtain here a result that is quite contrary to our intuition: bond
B, although having half of A's maturity (20 years as opposed to 40 years),
has a higher duration (12.34 years against 10.8 years) and therefore
exhibits more risk.
Value of bond for
i = 12% i = 14%
B
A
(i) $175.6 $147.4 hB
A
aB
A
= 16X1%
B
B
(i) $253.1 $205.22 hB
B
aB
B
= 18X9%
This is due of course to the nonmonotonicity property of duration with
maturity for under-par (low-coupon) bonds when maturity is very high,
which we explained in section 4.5.3.
4.10. dVaV is a random variable that is itself a weighted sum of two
random variables
~
A and
~
B,
~
AIdi and
~
BIdeae. The weights are
: I
D
1i
and = 1. We thus have
dVaV I:
~
A
~
B
Applying the formula for the variance of a weighted sum of dependent
random variables,
~
A and
~
B, we have
VAR(dVaV) = :
2
VAR(
~
A)
2
VAR(
~
B) 2:COV(
~
AY
~
B)
which establishes equation (18).
Chapter 5
5.1. As a check, you should obtain, for a 4% coupon and 20-year matu-
rity bond, a relative bias of 3.14% when rates of interest are at 10%. For
the same bond, the prot at delivery is $1.525. The graphs of all relative
biases and prots are given in gures 1a and 1b.
5.2. As a check, when rates of interest are at 6%, for a 4% coupon and 20-
year maturity bond, the relative bias is 3.26%. For the same bond, the
prot at delivery is $2.506. The graphs of all relative biases and prots
are given in gures 2a and 2b.
0 20 40 60 80 100
30
20
10
0
10
20
30
40
R
e
l
a
t
i
v
e

b
i
a
s

(
i
n

%
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 1a
Relative bias of conversion factor in T-bond futures contract, with i F10%
0 20 40 60 80 100
20
10
0
10
20
30
40
P
r
o
f
i
t

(
$
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 1b
Prot from delivery with T-bond futures contract; F(T ) F82X84; i F10%
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
0 20 40 60 80 100
15
10
5
0
5
10
R
e
l
a
t
i
v
e

b
i
a
s

(
i
n

%
)
Figure 2a
Relative bias of conversion factor in T-bond futures contract, with i F6%
0 20 40 60 80 100
15
10
5
0
5
10
15
P
r
o
f
i
t

(
$
)
2% coupon
4% coupon
6% coupon
8% coupon
12% coupon
Figure 2b
Prot from delivery with T-bond futures contract; F(T) F123.12; i F6%
Chapter 6
6.1. Using equation (5) in chapter 3, we can replace i
1
with i
0
. We then
get B
1
= B
0
, and therefore r
H
= i
0
for any H. Thus, (i
0
Y i
0
) is a xed point
for all curves r
H
(i).
6.2. Indeed, that kind of problem could not be tackled simply by the
intuitive reasoning we used in section 6.1 since it requires more rened
analysis.
What we want to show is that for any bond, either under- or above-par,
the graph pictured in gure 6.1 is correct. This amounts to showing the
following relationship:
dr
H
(i
0
)
di
E0 if and only if HED
In other words, the horizon rate of return has a positive (zero, negative)
slope at xed point (i
0
Y i
0
) if and only if the horizon H is larger than
(equal to, smaller than) duration.
Recall our formula (5) of chapter 3 for r
H
:
r
H
=
B(i)
B
0
_ _
1aH
(1 i) 1
Let us now study, in linear approximation, the relative rate of increase
of the dollar horizon rate of return 1 r
H
, that is,
1
1r
H
d log(1r
H
)
di
. We have
1 r
H
=
B(i)
B
0
_ _
1aH
(1 i) = (1aB
0
)
1aH
[B(i)[
1aH
(1 i) (1)
Denoting the constant
1
H
log(1aB
0
) by k, we have
log(1 r
H
) = k
1
H
log B(i) log(1 i) (2)
Furthermore, the derivative of log(1 r
H
) with respect to i can be written
as
d log(1 r
H
)
di
=
1
1 r
H
d(1 r
H
)
di
=
1
1 r
H
dr
H
di
(3)
and is equal, at point i
0
, to
1
1 r
H
(i
0
)
dr
H
(i
0
)
di
=
1
H
1
B(i
0
)
dB(i
0
)
di

1
1 i
0
(4)
Since 1 r
H
(i
0
) can be assumed to be always positive (because
r
H
(i
0
) b 1),
dr
H
(i
0
)
di
and the above expression (4) are always positive
(equal to zero, negative) if and only if
1
H
B
/
(i
0
)
B(i
0
)
E
1
1 i
0
(5)
or
(1 i
0
)
B(i
0
)
B
/
(i
0
)
B(i
0
)
DH (6)
(Note the changes in the inequality signs.) In the left-hand side of the set
of inequalities (6) we recognize the duration of the bond whatever its
coupon rate (and therefore its initial price) may be. Therefore, we can
write
dr
H
(i
0
)
di
E0 if and only if HED
Thus the intuitive reasoning that is valid for a bond at par can be
extended to above- or below-par bonds through analytical reasoning.
6.3. The ``variance of the bond'' means the dispersion of the bond's
times of payments around its duration and is measured by S =
T
t=1
(t D)
2
[c
t
(1 i)
t
aB(i)[, as we dened it in section 4.5.2.
6.4. Consider rst the left-hand side of (9) in this chapter. We have
d
2
ln F
H
di
2
=
d
di
d ln F
H
di
_ _
=
d
di
1
F
H
dF
H
di
_ _
; in the bracketed term, the units of F
H
cancel out, and therefore the remaining units in this term are those of 1ai,
which are years. Therefore,
d
di
1
F
H
dF
H
di
_ _
is in years/(1/year) = years
2
X
This property is conrmed both by the right-hand side of (9), which is
in years/(1/year) = years
2
(since (1 i) is unitless) and by the last equa-
tion of section 6.2, which shows that d
2
ln F
H
adi
2
= Sa(1 i)
2
, where S
is expressed in years
2
X
Chapter 7
7.1. There are at least two ways of calculating convexity; they rely on
either formula (2) or formula (19). Both formulas are open-form. A
spreadsheet is quite appropriate for such calculations, and it may be
advisable, for checking purposes, to use both equations (2) and (19). The
result is 62.79 (years
2
).
7.2. The explanation of this apparently nonintuitive result is as follows:
the bond's value curve is a sum of curves of the hyperbola type, which all
tend toward 0 when i increases indenitely, and therefore this curve has
an asymptote which is the horizontal i axis. On the other hand, the qua-
dratic approximation is a decreasing parabola; it will go through a mini-
mum and then start increasing toward innity. It is therefore quite normal
that at some point it will dominate both the bond's value curve and its
tangent.
7.3. The key assumption we make when calculating the bond portfolio's
convexity (or duration, or dispersion, for that matter) is that for each
bond the rate of return to maturity is the same.
7.4. After equating (17) to (1 i)S, where S is the dispersion of the
bond, remember that (1 i)B
/
aB is equal to duration D and replace
that expression in the bracketed term of (17). Then factor out 1aB from
the bracketed term and multiply throughout by 1 to obtain (18).
Chapter 8
8.1. There are, in fact, many ways to perform the calculations of table
8.1. The most natural one would be to determine closed-form formulas
both for duration and convexity. For duration, this closed form is given
in equation (11) of section 4.5.1. However, the calculation of the second
derivative of B(i), necessary for the determination of convexity, turns out
to be a very cumbersome exercise (its expression is spread out on three
lines), and writing a program to do this is no less of a chore. This is why
we hinted in this exercise to choose another route, founded upon the
approximations
D =
1 i
B(i)
dB
di
e
1 i
B(i)
hB
hi
(7)
and
C =
1
B(i)
d
2
B
di
2
e
1
B(i)
h
hi
hB
hi
_ _
(8)
which can be rendered as close to the exact value as we choose, simply by
considering suciently small, nite increases in i, hi.
It turns out that there are various ways of performing such calculations.
Quite an appropriate one would be to choose a rst increase of i equal to
hi = 10
5
= 0X00001, for instance. However, we can obtain better preci-
sion in the calculations by choosing the following method. Call
i
0
= 0X09
i
1
= 0X09 10
5
= 0X09001
i
2
= 0X09 10
5
= 0X08999
To each of these values of i correspond the following values of B(i) for
a bond at par (with coupon 9%) and maturity 10 years (easily determined
through the closed-form formula (13) of chapter 2):
i B(i)
i
0
= 0X09 B(i
0
) = 100
i
1
= 0X09001 B(i
1
) = 99X99358
i
2
= 0X08999 B(i
2
) = 100X0064
Thus, the rst derivative can very well be approximated, at point i
0
,
by
B
/
(i
0
) =
dB
di
(i
0
) e
B(i
1
) B(i
2
)
i
1
i
2
(instead of considering simply
B(i
1
)B(i
0
)
i
1
i
2
or
B(i
0
)B(i
2
)
i
0
i
2
, both of which would
have entailed more error than the rst formula; a simple diagram can
show nicely why this will generally be so).
The second derivativeto be used in the calculation of convexity
can easily be determined through the following approximation. Denoting
i
1
i
0
= i
0
i
2
= hi, we have
B
//
(i
0
) =
d
2
B
di
2
(i
0
) e
h
hi
hB
hi
(i
0
)
_ _
=
[B(i
1
) B(i
0
)[ [B(i
0
) B(i
2
)[
(i
1
i
0
)
2
These expressions are very handy to program and yield excellent
results, provided hi is suciently small. (Note that this is precisely the
way some nancial services determine duration and convexity for practi-
tioners.) It is quite remarkable that the results
D =
1 i
0
B(i
0
)
B
/
(i
0
) = 6X9952 years
and
C =
B
//
(i
0
)
B
/
(i
0
)
= 56X4962 (years
2
)
can be obtained with more than four decimals (ve, in fact) by using an
increase of one basis point for the rate of interest (hi = 0X0001), which
corresponds to two basis points for i
1
i
2
.
The gures in table 8.3 (horizon rates of return) do not give rise to any
special problem and are straightforward to establish.
8.2. The results contained in tables 8.78.10 are quite straightforward to
establish using the corresponding closed-form formulas.
a. Although the durations of both bonds are very close (about 7.9 years),
their convexities are signicantly dierent:
1
B
A
dB
2
A
di
2
= C
A
= 76X9 (years
2
)
1
B
B
dB
2
B
di
2
= C
B
= 67X2 (years
2
)
We now show why the higher convexity of B makes it preferable to A.
b. The rates of return at horizon H = D years are the following, if i
moves from i = 12% either to 10% or to 14%:
i = 10% i = 12% i = 14%
r
A
H=D
12.06% 12% 12.06%
r
H=D
12.03% 12% 12.03%
In either case, the higher convexity of A makes A more attractive than
B in a defensive management style.
c. The rate of return at horizon H = 1 year:
i = 10% i = 12% i = 14%
r
A
H=1
27.3% 12% 0.4%
r
H=D
27.1% 12% 0.7%
The results under (b) and (c) both conrm that bond A is to be pre-
ferred to bond B.
Chapter 9
9.1. See the table at the end of section 9.1.
9.2. To check your answer, replace the increasing spot interest rate
structure you have chosen with the rst line of table 9.1.
9.3. To check your answer, replace your hump-shaped spot interest rate
structure with the rst line of table 9.2.
Chapter 10
10.1. You have to use Leibniz's formula of the derivative of an integral
with respect to a parameter. Let : be such a parameter, and a denite
integral I = I(:), depending on : for these reasons: the function of x to be
integrated, f , depends on x and : and is written as f = f (xY :); further-
more, both integration bounds, a and b, depend on :. We thus have
I(:) =
_
b(:)
a(:)
f (xY :) dx
The Leibniz formula is the following:
I
/
(:) I
dI
d:
(:) =
_
b(:)
a(:)
f (xY :)
:
dx
db(:)
d:
f [b(:)Y :[
da(:)
d:
f [a(:)Y :[
(The demonstration of this formula can be found in any analysis text;
however, you should convince yourself that the formula makes sense by
drawing a simple diagram showing the exact increase in the denitive
integral hI entailed by a nite increase h: of parameter : and then con-
sidering the limit lim
h:0
hI
h:
= I
/
(:).)
In equation (6) of section 10.1, we have a special case of such an inte-
gral: only the function to be integrated depends on parameter :; the inte-
gration bounds are independent of :. Thus only the rst part of Leibniz's
formula applies to our problem. Taking the partial derivative of the
function to be integrated in (6) with respect to : and the denite integral
of this partial derivative from 0 to T yields the right-hand side of (8).
10.2. Our rst task is to dene in continuous time the horizon rate of
return for our investor when the spot rate structure
~
i receives a parallel
displacement :. In a way that extends what we did in section 6.2 (chapter
6), we can dene this horizon rate of return r
H
such that it transforms the
bond's initial value B(
~
i ) into a future value F
H
when the structure
~
i has
undergone a shock : immediately after the bond's purchase. This future
value F
H
(
~
i :) is given by equation (11) in section 10.3. Thus, r
H
is
dened by the following relationship:
B(
~
i )e
r
H
H
= F
H
= B(
~
i :)e
[i(0Y H):[H
(9)
Dividing equation (9) by B(
~
i ), taking the logarithm on both sides, and
nally dividing by H, yields
r
H
=
1
H
log
B(
~
i :)
B(
~
i )
_ _
i(0Y H) : (10)
We can verify that in the case where no shock takes place (: = 0), the
horizon rate of return is equal to i(0Y H), as it should be.
We have now a function r
H
= r
H
(:) in (r
H
Y :) space that we can try to
minimize, in a way similar to what we did in section 6.2 (chapter 6). Con-
sider equation (10) and perform in reverse order the operations described
in the paragraph preceding it. Minimizing the horizon rate of return is
thus equivalent to minimizing the future value of the investment, and
we can then continue our demonstration of the immunization theorem
with sections 10.4 and 10.5.
Chapter 11
11.1. Applying formula (3) of section 11.1 with m = 6, hz
1
= 0X5 years,
i
(1)
1
= 6% per year, we get i
(m)
1
=
6
0X5
[1 0X06(1a2)[
1a6
1 = 5X926% per
year, which is smaller than 6%, as it should be.
11.2. Applying (7), we have i
(y)
1
=
log[10X06(1a2)[
1a2
= 5X912% per year, which
is the smallest interest rate that would yield the same return to the lender
(or cost the same amount to the borrower).
11.3. a. From R(0Y T) = 0X005T
2
0X2T 4 (expressed in percent per
year), the spot rate is indeed 5.5% per year for a horizon of 10 years.
b. The area of the rectangle is (5.5% per year 10 years) = 0.55 (a pure
number). This area can be veried to be the integral (the sum) of the
innitely large number of forward rates from 0 to 10,
_
10
0
f (0Y T) dT
=
_
10
0
(0X015T
2
0X4T 4) dT = [0X005T
3
0X2T
2
4T[
10
0
= 55% =
0X55.
c. Note that this must be a pure number (as observed above) because in
the denite integral you are considering a sum of innitesimal elements
f (0Y T) dT, whose dimension is indeed (1/time) time, a pure number as it
should be.
d. The economic (or nancial) interpretation of this areaexpressed as
the pure number 0.55, or 55%is the continuously compounded rate of
interest per 10-year period that applies to a loan made at time 0 and
reimbursed 10 years later (or, equivalently, 5.5% per year). Indeed, we
have
C
10
= C
0
e
R(0Y 10)10
= C
0
e
(0X055)10
= C
0
e
0X55
and therefore
ln(C
10
aC
0
) = 0X55 = 55%
e. In fact, you could have dispensed with any calculation: ln(C
10
aC
0
) =
0X055 10, the continuously compounded rate of return over a 10-year
period, must be equal to the denite integral of the forward rates
_
10
0
f (0Y T) dT you have determined, both R(0Y 10) = 5X5% per year and
T = 10 years having been given to you in (a). But it is denitely a smart
idea to verify at least once those important properties.
Chapter 12
12.1. You have to perform the following integration: R(T) =
1
T
[
_
T
0
f (u) du[; f (u) is obtained through equation (14). You just have to be
careful that in (14), T is now transformed into an integration variable.
Therefore, you have to calculate
R(T) =
1
T
_
T
0
f (u) du
_ _
=
1
T
_
T
0
0

1
e
ua
1
1
ue
ua
1
_ _
du
_ _
which does not present any diculty. Integrating once by parts the last
term yields (15).
12.2. First, write down a modied Hamiltonian corresponding to (14).
Here x replaces t. The rest of the demonstration is given in Appendix
12.A.2.
12.3. A rst integration of y
(4)
(x) leads to
y
///
(x) = c
1
where c
1
is a constant. A second integration enables us to write
y
//
(x) = c
1
x c
2
and so forth:
y
/
=
c
1
2
x
2
c
2
x c
3
y =
c
1
6
x
3
c
2
2
x
2
c
3
x c
4
which amounts to a third-degree polynomial.
Chapter 13
13.1. Using the fact that the normal density of the unit normal variable Z
is (2)
1a2
e
(1a2)z
2
, the moment-generating function (with parameter p) of
the unit normal is
Z
(p) = E(e
pZ
) =
_
y
y
e
pz
1

2
e
z
2
a2
dz =
_
y
y
1

2
e
z
2
2
pz
dz (11)
Observe, in the right-hand side of (11), that the exponent of e can be
written as
z
2
2
pz =
1
2
(z
2
2pz p
2
) p
2
a2 =
1
2
(z p)
2
p
2
a2.
(This operation is called ``completing the square'' because you observe
that the exponent of e is ``nearly'' the perfect square of an ane function
of z, z p.) Making now the change of variable z pa2 = y, the moment-
generating function of Z can be written as
Z
(p) = e
p
2
a2
_
y
y
1

2
e
1
2
y
2
dy (12)
The integral in the right-hand side of (12) is just 1 (the integral of the unit
normal density from yto y). Hence the moment-generating function
of Z is simply
Z
(p) = e
p
2
a2
(13)
The moment-generating function of the normal U = oZ@
N( Y o
2
) is
U
(z) = E(e
zU
) = E(e
z zoZ
) = e
z
E(e
zoZ
)
Now in order to evaluate E(e
zoZ
), it suces to observe that this
expression is none other than the moment-generating function of Z,
where zo now plays the role of the parameter p. We then have
E(e
zoZ
) =
Z
(zo) = e
z
2
o
2
a2
and hence the moment-generating function of U @ N( Y o
2
) is
U
(z) = E(e
zu
) = E(e
z zoZ
) = e
z z
2
o
2
a2
This result will be of great importance in this and subsequent
chapters.
13.2. Applying equations (5), (8), and (9) of chapter 13, we get
E(R
t1Y t
) = e

o
2
2
1 = 10X52%
VAR(R
t1Y t
) = e
2 o
2
[e
o
2
1[ = 4X98%
o(R
t1Y t
) = e

o
2
2
[e
o
2
1[
1a2
= 22X33%
13.3. We have to determine
Prob(R
t1Y t
` E(R
t1Y t
))
This probability is equal to
Prob(1 R
tt
= X
t1Y t
` 1 E(R
t1Y t
) = E(X
t1Y t
))
= Prob(X
t1Y t
` e

o
2
2
) = Prob log X
t1Y t
`
o
2
2
_ _
Since log X
t1Y t
@ N( Y o
2
), log X
t1Y t
= oZ, Z@ N(0Y 1), where Z
denotes Z
t1Y t
for short. Hence this probability is equal to
Prob oZ `
o
2
2
_ _
= Prob Z `
o
2
_ _
This result is quite striking in its simplicity.
13.4. Denote f (z) as the probability density of the unit normal variable.
With o
2
= 4%, using the result of question 13.3, the probability that
R
t1Y t
is smaller than E(R
t1Y t
) is
Prob(R
t1Y t
` E(R
t1Y t
) = Prob(Z ` 0X01) =
_
0X1
y
f (z) dz = 0X5398
13.5. Prob (R
t1Y t
` E(R
t1Y t
)) as a function of o is as follows:
o Prob(R
t1Y t
` E(R
t
1
1Y t
)) =
_
oa2
y
f (z) dz
0.05 0.51
0.1 0.52
0.15 0.53
0.2 0.54
0.25 0.55
0.3 0.56
0.35 0.57
This probability, within this range of o values, is pretty much an ane
function of o.
13.6. A rst method would be to apply equations (16) and (22); we would
get
E(R
0Y n
) = e

o
2
2n
1 = 8X55%
o(R
0Y n
) = e

o
2
2n
(e
o
2
n
1)
1a2
= 6X87%
A second method would be to apply equations (25) and (28), using
E I1 E(R
t1Y t
) and V IVAR(R
t1Y t
) in them, as obtained in 13.2.
(Note, however, that these are the data that are more commonly used by
practitioners.) Applying (4) and (8), respectively, E and V turn out to be
E = 1X10517 and V = 0X04985. We get
E(R
0Y n
) = E (1 VaE
2
)
1
2
(
1
n
1)
1 = 8X55%
o(R
0Y n
) = [1 E(R
0Y n
)[[(1 VaE
2
)
1an
1)
1a2
= 6X87%
as they should.
13.7. a. We want to determine
Prob(R
0Y n
` E(R
0Y n
))
From equation (12) of this chapter, we know that
R
0Y n
= X
1an
0Y n
1
Therefore, 1 R
0Y n
= X
1an
0Y n
, and prob(R
0Y n
` E(R
0Y n
)) is equal to
Prob(1 R
0Y n
` 1 E(R
0Y n
) = Prob(X
1an
0Y n
` E(X
1an
0
))
= Prob
1
n
log X
0Y n
` log E(X
1an
0
)
_ _
From (13), we know that log X
0Y n
= n

n
oZ, Z@ N(0Y 1) and from

(15), E(X
1an
0
) = e

o
2
2n
. Therefore, this probability is equal to
Prob
o

n
Z `
o
2
2n
_ _
As in question 13.3, cancels out. We are left with the simple, and
important, result:
Prob(R
0Y n
` E(R
0Y n
)) = Prob Z `
o
2

n
_ _
b. As a check, if we plug n = 1 into this result, we get the same result
obtained in exercise 13.3: prob(R
0Y 1
` E(R
0Y 1
)) = prob Z ` oa2.
c. When n tends toward y, we have
Prob(R
0Y y
` E(R
0Y y
)) = Prob(Z ` 0) = 50%
13.8. You can work either in terms of , o
2
or in terms of E, V.
a. In terms of = 0X08 and o
2
= 0X04, apply the left-hand side of equa-
tion (29):
E(R
0Y y
) = e
1 = 8X33%
b. In terms of E, V = 1X1052, 4X98% as obtained in question 13.2, apply
the right-hand side of equation (29):
E(R
0Y y
) =
E
2
(E
2
V)
1a2
1 = 8X33%
Chapter 14
14.1. It implies that the average of all forward rates f (0Y ) for = 0 to
= 5,
_
5
0
f (0Y ) d a5, is equal to 5.069%.
14.2. No, because it implies only an increase in the average of forward
rates, not necessarily an increase in all forward rates.
14.3. These directional coecients are dimensionlessthey are pure
numbers. There are at least two ways of seeing this.
First, consider equation (7). Its left-hand side, dB, is in dollars; so
must be its right-hand side. dz is in (1/time) units; e
r
0
0Y t
t
is unitless; so
c
t
(t)e
r
0
0Y t
t
dz is in dollars. Hence a
t
must be dimensionless.
Second, consider equation (8). Its left-hand side is in 1/(1/years) =
years. On the other hand, in the right-hand side of (8), c
t
e
r
0Y t
t
aB is in
($/$), or a pure number. Since t is in years, a
t
must be a pure dimension-
less number.
14.4. a. The exact relative change in the bond's price is
hB
B
=
B
1
0
B
0
0
B
0
0
=
1112X464 1118X254
1118X254
= 0X518%
b. The directional duration of the bond is equal to D
a
= 5X192 years.
c. The relative change in the bond's price in linear approximation is given
by
dB
B
0
0
= D
a
d. Using 5 decimals for dBaB and hBaB, the relative error you are
introducing by using this linear approximation is
relative error =
(dBaB) (hBaB)
hBaB
= 0X29%
Indeed, you could have gured out what the orders of magnitude of the
answers to all questions (a), (b), (c), and (d) would be by reasoning in the
following way.
Consider the directional vector a in this case. It has the following
properties: it lies between the directional vector used in the example
treated in section 14.3, and the unit vector corresponding to a parallel
shift of the continuously compounded term structure, but it is denitely
closer (in many senses) to the directional vector leading to a nonparallel
displacement. We can conclude that
a. the exact relative change in the bond's price will be slightly larger in
absolute value than in the nonparallel displacement case we treated in this
chapter (in algebraic value, 0.518% against 0.444%);
b. the directional duration of the bond is slightly higher (5.192 years
against 4.440 years);
c. the relative change in the bond's price in linear approximation should
be slightly higher in absolute value (in algebraic value: 0.519% instead
of 0.444%);
d. the relative error made by using the linear approximation should prove
to be slightly higher (0.29% against 0.25%).
Table 14a summarizes the results of this project.
Chapter 15
15.1. The polynomial corresponding to the new spot rate curve is given
by
R
+
(t) = 0X06 0X00218t 0X000083t
2
(1X2)10
6
t
3
15.2. The optimal (immunizing) portfolio is the one given in this
chapter.
15.3. The seven-year horizon portfolio value corresponding to the new
structure is 103X3544e
(0X04836)7
= $144X99. It is even higher than the initial
portfolio's value ($144.72).
Chapter 16
16.1. In (6a), factor out e
iht
to get
V
0
= e
iht
f (S
u
) f (S
d
)
S
u
S
d
S
0
e
iht
f (S
u
) S
u
f (S
u
) f (S
d
)
S
u
S
d
_ _
(6a
/
)
In the bracketed term of (6a
/
), factor out f (S
u
) and f (S
d
):
V
0
= e
iht
f (S
u
)
e
iht
S
0
S
u
S
d
1
S
u
S
u
S
d
_ _
f (S
d
)
S
u
S
0
e
iht
S
u
S
d
_ _ _ _
(6a
//
)
433
T
a
b
l
e
1
4
a
N
o
n
p
a
r
a
l
l
e
l
d
i
s
p
l
a
c
e
m
e
n
t
o
f
t
h
e
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
c
u
r
v
e
T
e
r
m
o
f
p
a
y
m
e
n
t
C
a
s
h
o
w
I
n
i
t
i
a
l
s
p
o
t
r
a
t
e
(
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
)
I
n
i
t
i
a
l
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
D
i
r
e
c
t
i
o
n
a
l
c
o
e
c
i
e
n
t
v
e
c
t
o
r
a
C
a
l
c
u
l
a
t
i
o
n
o
f
d
i
r
e
c
t
i
o
n
a
l
t
e
r
m
s
I
n
c
r
e
a
s
e
i
n
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
N
e
w
c
o
n
t
i
n
u
o
u
s
l
y
c
o
m
p
o
u
n
d
e
d
s
p
o
t
r
a
t
e
N
e
w
d
i
s
c
o
u
n
t
e
d
c
a
s
h
o
w
s
t
C
t
l
n
(
1
i
0
Y
t
)
=
r
0 0
Y
t
c
t
e
r
0 0
Y
t
t
a
t
t
a
t
c
t
e
r
0 0
Y
t
t
a
B
0 0
a
t
d
z
r
1 0
Y
t
=
r
0 0
Y
t
a
t
d
z
c
t
e
r
1 0
Y
t
t
1
7
0
0
.
0
4
4
0
1
7
6
6
.
9
8
6
1
0
.
0
5
9
9
0
.
0
0
1
0
.
0
4
5
0
1
7
6
6
.
9
1
9
2
7
0
0
.
0
4
6
4
0
6
6
3
.
7
9
5
0
.
9
5
0
.
1
0
8
4
0
.
0
0
0
9
5
0
.
0
4
7
3
5
6
6
3
.
6
7
4
3
7
0
0
.
0
4
8
3
1
4
6
0
.
5
5
5
0
.
9
0
.
1
4
6
2
0
.
0
0
0
9
0
.
0
4
9
2
1
4
6
0
.
3
9
2
4
7
0
0
.
0
4
9
7
4
2
5
7
.
3
7
0
0
.
8
5
0
.
1
7
4
4
0
.
0
0
0
8
5
0
.
0
5
0
5
9
2
5
7
.
1
7
6
5
7
0
0
.
0
5
0
6
9
3
5
4
.
3
2
7
0
.
8
0
.
1
9
4
3
0
.
0
0
0
8
0
.
0
5
1
4
9
3
5
4
.
1
8
1
6
7
0
0
.
0
5
1
6
4
3
5
1
.
3
4
9
0
.
7
5
0
.
2
0
6
6
0
.
0
0
0
7
5
0
.
0
5
2
3
9
3
5
1
.
1
1
8
7
7
0
0
.
0
5
2
5
9
2
4
8
.
4
4
1
0
.
7
0
.
2
1
2
3
0
.
0
0
0
7
0
.
0
5
3
2
9
2
4
8
.
2
0
4
8
7
0
0
.
0
5
3
0
6
7
4
5
.
7
8
5
0
.
6
8
0
.
2
2
2
7
0
.
0
0
0
6
8
0
.
0
5
3
7
4
7
4
5
.
5
3
7
9
7
0
0
.
0
5
3
5
4
1
4
3
.
2
3
4
0
.
6
5
0
.
2
2
6
2
0
.
0
0
0
6
5
0
.
0
5
4
1
9
1
4
2
.
9
8
2
1
0
1
0
7
0
0
.
0
5
3
5
4
1
6
2
6
.
4
1
1
0
.
6
5
3
.
6
4
1
1
0
.
0
0
0
6
5
0
.
0
5
4
1
9
1
6
2
2
.
9
7
5
B
0 0
=
1
1
1
8
X
2
5
4
D
i
r
e
c
t
i
o
n
a
l
d
u
r
a
t
i
o
n
:
D
a
=
5
X
1
9
2
2
B
1 0
=
1
1
1
2
X
4
6
4
Simplifying now the rst bracketed term of (6a
//
), we get equation (7) in
the text.
16.2. That an innity of derivative values f (S
u
) and f (S
d
) leading to the
same value V
0
corresponds to one set of values iY S
0
Y S
u
, and S
d
comes
from the fact that V
0
is determined by the single equation (7), whose right-
hand side is a function of the two variables f (S
u
) and f (S
d
), given
i, S
0
Y S
u
, and S
d
.
16.3. Solve equation (7) in f (S
d
) with the following parameters:
4X8 = e
0X11
[4X6(0X38) f (S
d
)0X61[
You get f (S
d
) = 11X7, which can be checked with the height of point
D
//
in gure 16.1.
16.4. You need the following information: rst, the interest rate pertain-
ing to the time span from the second to the third period and all possible
values of the underlying asset at each note, which will enable you to de-
termine the martingale probabilities Q; second, the conditional proba-
bilities of reaching each node of period 3.
16.5. You simply have to develop the rst part of equation (26), not for-
getting that a I
1
o
( o
2
a2 r).
Chapter 17
17.1. You rst have to derive the current rate r(t) from the forward
rate process given in integral form by equation (2), and then evaluate
exp[
_
t
0
r(s) ds[.
17.2. See the detailed required steps in appendix 17.A.1.
17.3. You can retrace the various steps described in part 1 of appendix
17.A.2. Note the crucial point that when dt becomes innitesimally small,
(dW
t
)
2
ceases to become a random variable and becomes equal to dt.
17.4. This is a direct application of Ito 's lemma (see appendix 17.A.2.).
17.5. You just need to calculate oe
z(Tt)
_
T
t
oe
z(ut)
du; after the neces-
sary simplications, you will obtain the drift term :(t).
17.6. The following three denite integrations are needed:
a. the one you have performed in question 17.5,
b. the integration of the dierential df (equation (20)), and
c. the integration of exp[
_
T
t
f (tY u) du[, which leads to the value of
B(tY T) (equation (22)). The integrations are fortunately straightforward
for the nonrandom terms.
Chapter 18
18.1. The HJM portfolios and the classical portfolio can be easily calcu-
lated by inverting the matrix M:
M =
B
a
B
b
B
c
X
a
B
a
X
b
B
b
X
c
B
c
X
2
a
B
a
X
2
b
B
b
X
2
c
B
c
_
_
_
_
and multiplying the inverted matrix M
1
by the vector (B; XB;
X
2
B)
/
. The durations are straightforward to determine, and the results
are summarized in the table 18a.
Table 18a
Hedging portfolios and their durations*
Number of
bonds in
hedging portfolio
Classical
portfolio
z = 0 z = 0X075 z = 0X125 z = 0X175 z = 0X225
n
a
0.325 0.313 0.305 0.297 0.290
n
b
1.025 1.064 1.091 1.119 1.147
n
c
0.351 0.378 0.397 0.417 0.437
Portfolio's duration 1 1 0.999 0.997 0.996
*Rounded to the third decimal place
18.2. The immunization portfolios are given in table 18b.
Table 18b
Immunization results (summary)
Heath-Jarrow-Morton portfolios
Number
of bonds
Classical
portfolio
z = 0 z = 0X1 z = 0X15 z = 0X2 z = 0X25 z = 0X3
N
a
1.148 0.977 0.957 0.966 0.976 0.967
N
b
1.996 1.766 1.733 1.746 1.776 1.791
N
c
0.117 0.160 0.169 0.166 0.152 0.135
Duration
(years)
8.023 8.061 8.083 8.076 8.034 7.965
Those results indicate rst that the components of the vector (N
A
Y N
B
Y
N
C
)the components of the replicating portfolioare not monotonic
functions of the reduction volatility factor z. Also, we notice that when z
increases from z = 0 to z = 0X3, the duration of the immunizing port-
folio rst increases above the classical portfolio's duration (8.023), then
decreases to reach a value below that duration. This suggests that there
must exist at least one z value for which both durations are the same. This
proves to be true, as table 18c shows.
Table 18c
Immunization results for l values between l0.258 and lF0.26, and for classical portfolio
(lF0)
Heath-Jarrow-Morton portfolios
Number of bonds
Classical
portfolio
z = 0 z = 0X258 z = 0X259 z = 0X26
N
a
1.148 0.976 0.976 0.976
N
b
1.996 1.780 1.780 1.781
N
c
0.117 0.150 0.149 0.149
Duration (years) 8.023 8.024 8.023 8.022
For z = 0X259, the corresponding Heath-Jarrow-Morton portfolio,
although somewhat dierent from the classical portfolio, exhibits the
same duration (8.023 years).
FURTHER READING
Ferdinand, King of Navarre. How well he's read,
to reason against reading!
A complete bibliography on bonds would take more than one volume,
since nearly every professional nancial journal has at least one paper on
the subject in each of its issues. For example, the remarkable book by
Marek Musiela and Marek Rutkowski, Martingale Methods in Financial
Modelling (Cambridge, U.K.: Cambridge University Press, 1998)written
from a methodological point of viewlists more than one thousand recent
references rather closely related to our subject, and many are of a highly
technical nature.
We will therefore limit ourselves here to a number of books and articles
that we nd essential, either for the new methodology they oer or for the
historical development of the subject. Also, some books contain a lot of
institutional information, such as auctions, special kinds of bonds, and so
forth.
Many of the issues discussed in this book have been treated in a very
lucid way by David Luenberger in his text Investment Science (Oxford:
Oxford University Press, 1998). At about the same level of technicality as
Luenberger's text is the now classic Options, Futures and Derivative
Securities by John Hull (Englewood Clis, N.J.: Prentice Hall, 1993). A
detailed descriptive and institutional point of view is taken both in the
recent book by Suresh Sundaresan, Fixed Income Markets and Their
Derivatives (Cincinnati, Ohio: South-Western, 1997), and in Frank
Fabozzi's book The Handbook of Treasury Securities (Probus, 1986). In-
terest rate derivatives are thoroughly covered in Derivative Securities by
Robert Jarrow and Stuart Turnbull (Cincinnati, Ohio: South-Western,
1996). A deep account of the implications of a given spot interest struc-
ture will be found in the thorough study by Robert Shiller, ``The Term
Structure of Interest Rates,'' in Benjamin Friedman and Frank Hahn
(eds.), Handbook of Monetary Economics (North Holland, 1989).
From a methodological point of view, we feel that the recent book by
Martin Baxter and Andrew Rennie, An Introduction to Derivative Pricing,
(Cambridge, U.K.: Cambridge University Press, 1998) is a must for any-
one not familiar with concepts such as probability measure change and
the Cameron-Martin-Girsanov theorem. We have made ample use of it
in section 13.1 (chapter 13) of this book. It is remarkable in its clarity. At
an advanced level, the book by Marek Musiela and Marek Rutkowski
cited above oers the reader very current insights.
In these past twenty years, research has organized itself rst in con-
tinuous time models, which were followed by discrete time analysis, and
returned nally to continuous models. We now describe briey some im-
portant contributions in this area.
Research began with arbitrage models in continuous time with one or
more state variables. In general, this type of model is based on the speci-
cation of diusion processes governing interest rates and the denition
of risk premia. The rst model in this vein was due to Vasicek (`Àn
Equilibrium Characterization of the Term Structure,'' Journal of Finan-
cial Economics, 5 (November 1977): 177188), who developed a stochas-
tic model using one state variable (the short-term rate). Two-state
variable models were introduced by S. Richard (`Àn Arbitrage Model of
the Term Structure of Interest Rates,'' Journal of Financial Economics, 6
(1978) and Cox, Ingersoll, and Ross (`Àn Intertemporal General Equi-
librium Model of Asset Prices,'' Econometrica, 53, no. 2 (1985): 363384),
where the second-state variable is the ination rate.
Nelson and Schaefer (``The Dynamics of the Term Structure and
Alternative Portfolio Immunization Strategies,'' in G. O. Bierwag, G.
Kaufman, and A. Toevs, eds., Innovation in Bond Portfolio Management:
Duration Analysis and Immunization (Greenwich, Conn.: JAI Press 1983))
and Schaefer and Schwartz (`À Two-Factor Model of the Term Struc-
ture: An Approximate Solution,'' Journal of Financial and Quantitative
Analysis, no. 4 (1984)) considered as state variables the spread dierence
between the long-term rate and the short-term rate. The contribution
found in Cox, Ingersoll, and Ross (`À Theory of the Term Structure of
Interest Rates,'' Econometrica, 53, no. 2 (1985): 385407) is a general
equilibrium model of considerable breadth.
In recent years, the tendency has been to develop the modeling of the
whole-term structure, following a suggestion made by T. S. Y. Ho and
S. B. Lee in their article ``Term Structure Movements and Pricing Interest
Rate Contingent Claims'' (Journal of Finance, December (1986): 1011
1029). This model, which rests also on arbitrage pricing, was recently
extended by Hull and White in `Òne-Factor Interest Rate Models and
the Valuation of Interest Rate Derivative Securities'' (Journal of Finan-
cial and Quantitative Analysts, June (1993): 235254). Another approach
consists of dening a process for the short rate; this was used by Black,
Derman, and Toy in `À One-Factor Model of Interest Rates and Its
438 Further Reading
Application to Treasury Bond Options'' (Financial Analysis Journal,
January (1990): 3339) and Hull and White in ``Pricing Interest Rate
Derivative Securities'' (Review of Financial Studies, 4, (1990): 573592).
To the best of our knowledge, the idea of developing the price of a
discount bond in a Taylor series is due to Donald R. Chambers and
Willard T. Carleton, `À Generalized Approach to Duration,'' Research
in Finance, 7 (1988): 163181. For immunization purposes, the model was
very successfully tested with real data by D. R. Chambers, W. T. Carleton,
and R. W. McEnally, `Ìmmunizing Default-Free Bond Portfolios with a
Duration Vector,'' Journal of Financial and Quantitative Analysis, 23,
no. 1 (March 1988): 89104.
As we have stressed in the introduction as well as in the last two chap-
ters of this book, a general and innovative approach to the problem of
modeling the term structure from a stochastic point of view was suggested
by David Heath, Robert Jarrow, and Andrew Morton in their now classic
paper, ``Bond Pricing and the Term Structure of Interest Rates: A New
Methodology for Contingent Claims Valuation'' (Econometrica, 60, no. 1
(January 1992): 77105). At an advanced level, the reader will certainly
be interested in the recent paper by Martin Baxter, ``General Interest Rate
Models and the Universality of HJM'' (in Mathematics of Derivative
Securities, M. A. H. Dempster and S. R. Pliska, eds., pp. 315335)
(Cambridge, U.K.: Cambridge University Press, 1997).
Recent advances in stochastic optimization of bond portfolios have
been carried out by a number of researchers, among them Marida Ber-
tocchi and Vittorio Moriggia (of the Department of Mathematics at the
University of Bergamo), Jitka Dupacova (Department of Probability and
Mathematical Statistics, University of Prague), and Stavros Zenios (Uni-
versity of Cyprus, and Decision Sciences Department, The Wharton
School, University of Pennsylvania). In particular, the reader should have
a close look at two papers published in W. Ziemba and J. Mulvey,
Worldwide Asset and Liability Modelling (Cambridge, U.K.: Cambridge
University Press, 1998): J. Dupacova, M. Bertocchi, and V. Moriggia,
``Postoptimality for Scenario Based Financial Planning Models with an
Application to Bond Portfolio Management'' (pp. 263285), and S.
Zenios, `Àsset and Liability Management under Uncertainty for Fixed
Income Securities'' (pp. 537560).
Also in the same vein, and of great interest, are the following studies:
M. Bertocchi, J. Dupacova, and V. Moriggia, ``Security of Bond Portfolios
Behavior with Respect to Random Movements in Yield Curve: A Simula-
439 Further Reading
tion Study,'' to appear in Annals of Operations Research (1999); J. Dupa-
cova and M. Bertocchi, ``Management of Bond Portfolios via Stochastic
ProgrammingPostoptimality and Sensitivity Analysis'' (in J. Dolezal
and J. Fidler, eds., System Modelling and Optimization, New York:
Chapman and Hall, 1995, 574582); and J. Dupacova and M. Bertocchi,
``From Data to Model and Back to Data: A Bond Portfolio Management
Problem'' (to appear in the European Journal of Operations Research).
Finally, the reader should also consult the following recent contributions:
A. Beltratti, A. Consiglio, and S. Zenios, ``Scenario Modelling for the
Management of International Bond Portfolios'' (Annals of Operations
Research, 85 (1999): 227247), and A. Consiglio and S. Zenios, `À Model
for Designing Callable Bonds and Its Solution Using Tabu Research''
(Journal of Economic Dynamics and Control, 21 (1997): 14451470). All
of these papers contain numerous references to related research.
The reader should also be aware of an important area of research that
deals with the optimization of portfolios carrying complex securities, such
as callable or putable bonds, mortgage-backed securities, and insurance
products. In order to tackle the problems raised, in particular, by cash
infusion and withdrawal and transaction costs for certain types of securi-
ties, the development of multiperiod dynamic models for portfolio man-
agement was called for. At an expository level, we refer the reader to R. S.
Hiller and C. Schaak, `À Classication of Structured Bond Portfolio
Modeling Techniques'' (Journal of Portfolio Management (1990): 3748).
For the problem of managing portfolios of mortgage-backed securities,
see S. A. Zenios, `À Model of Portfolio Management with Mortgage-
Backed Securities'' (Annals of Operations Research, 43 (1993): 337356),
as well as D. R. Carion et al., ``The Russel-Yasuda Kasai Model: An
Asset/Liability Model for a Japanese Insurance Company Using Multi-
stage Stochastic Programming'' (Interfaces 24(1) (1994): 2949). A suc-
cessful application to money management can be found in B. Golub, M.
Holmer, R. MacKendall, and S. Zenios, ``Stochastic Programming Models
for Money Management'' (European Journal of Operational Research, 85
(1995): 282296). At an advanced level, the reader will nd a textbook
treatment in Y. Censor and S. Zenios, Parallel Optimization: Theory,
Algorithms and Applications (New York: Oxford University Press, 1997).
An interesting new approach for various currency risk management
strategies was developed by A. Beltratti, A. Laurent, and S. Zenios in
``Scenario Modeling of Selective Hedging Strategies'' (The Wharton
School, University of Pennsylvania, Working Paper 99-15, 1999).
440 Further Reading
REFERENCES
Berowne. They are the ground, the books, the academes,
From whence doth spring the true Promethean re.
Loves' Labour's Lost
Petruchio. A herald, Kate? O put me in thy books!
The Taming of the Shrew
These references should be considered of course as nothing but a small
sample of the true Promethean re.
Adams, K. J., and D. R. Van Deventer. ``Fitting Yield Curves and Forward Rate Curves
with Maximum Smoothness.'' Journal of Fixed Income, June (1994): 5262.
Aitchinson, J., and J. A. C. Brown. The Lognormal Distribution. Cambridge, U.K.:
Cambridge University Press, 1973.
Anderson, N., and J. Sleath. ``New Estimates of the UK Real and Nominal Yield Curves.''
Bank of England Quarterly Bulletin, November (1999): 384392.
Arak, M., S. G. Laurie, and S. Ross. ``The Cheapest to Deliver Bond on the Treasury Bond
Futures Contract.'' In: Advances in Futures and Options Research, Vol. 1, Part B, pp. 4974.
Greenwich, Conn.: JAI Press, 1986.
Bachelier, L. ``Theorie de la speculation,'' Annales scientiques de l'Ecole normale superieure,
17 (1900): 2186. English translation in: The Random Character of Stock Market Prices. ed.
P. H. Cootner, pp. 1778. Cambridge, Mass.: MIT Press, 1964.
Baxter, M. ``General Interest Rate Models and the Universality of HJM,'' in Mathematics of
Derivative Securities, ed. M. A. H. Dempster and S. R. Pliska, pp. 315335. Cambridge,
U.K.: Cambridge University Press, 1997.
Baxter, M., and A. Rennie. An Introduction to Derivative Pricing. Cambridge, U.K.:
Cambridge University Press, 1998.
Beltratti, A., A. Consiglio, and S. Zenios. ``Scenario Modelling for the Management of
International Bond Portfolios.'' Annals of Operations Research, 85 (1999): 227247.
Beltratti, A., A. Laurent, and S. Zenios. ``Scenario Modeling of Selective Hedging Strat-
egies.'' Working Paper no. 99-15, Philadelphia Penna.: The Wharton School, University of
Pennsylvania, 1999.
Bertocchi, M., J. Dupacova, and V. Moriggia. ``Security of Bond Portfolios Behavior with
Respect to Random Movements in Yield Curve: a Simulation Study.'' In Annals of Oper-
ations Research, 1999.
Bierwag, G. Duration Analysis: Managing Interest Rate Risk. Cambridge, Mass.: Ballinger,
1987.
Bierwag, G., G. Kaufman, and A. Toevs, ``Recent Developments in Bond Portfolio Im-
munization Strategies,'' in Innovations in Bond Portfolio Management, ed. G. Kaufman,
G. Bierwag, and A. Toevs. Greenwich, Conn.: JAI Press, 1983.
Black, Derman, and Toy. `À One-Factor Model of Interest Rates and Its Application to
Treasury Bond Options.'' Financial Analysts Journal, January (1990): 3339.
Black, F. `Èstimating Expected Return.'' Financial Analysts Journal, SeptemberOctober
(1993): 3638.
Black, F., and M. Scholes. ``The Pricing of Options and Corporate Liabilities.'' Journal of
Political Economy, 81 (1973): 637654.
Blank, S., C. Carter, and B. Schmiesing. Futures and Options Markets. Englewood Clis,
N.J.: Prentice Hall, 1991, p. 384.
Bliss, Robert R. ``Testing Term Structure Estimation Methods.'' Advances in Futures and
Options Research, 9 (1997): 197231.
Bodie, Z., A. Kane, and A. B. Marcus. Investments. Burr Ridge, Ill.: Irwin, 1993.
de Boor, C. A Practical Guide to Splines. Springer-Verlag, 1978.
Bouchaud, J.-Ph., and M. Potters. Theorie des risques nanciers, Alea Saclay, Paris; to be
published as Theory of Financial Risk: from Statistical Physics to Risk Management. Cam-
bridge, U.K.: Cambridge University Press, 2000.
Brennan, M. J., and E. S. Schwartz. `Ànalyzing Convertible Bonds.'' Journal of Financial
and Quantitative Analysis, 15 (1980): 907929.
Brennan, M. J., and E. S. Schwartz. ``The Case for Convertibles.'' Journal of Applied
Corporate Finance, 1 (1977): 5564.
Cameron, R. H., and W. T. Martin. ``Transformations of Wiener Integrals under Trans-
lations.'' Annals of Mathematics, 45, no. 2 (April 1944): 386396.
Campbell, J., A. Lo, and A. MacKinlay. The Econometrics of Financial Markets. Princeton,
N.J.: Princeton University Press, 1997.
Carion, D. R., et al. ``The Russel-Yasuda Kasai Model: An Asset/Liability Model for a
Japanese Insurance Company Using Multistage Stochastic Programming.'' Interfaces, 24(1)
(1994): 2949.
Censor, Y., and S. Zenios. Parallel Optimization: Theory, Algorithms and Applications. New
York: Oxford University Press, 1997.
Chambers, D. R., and W. T. Carleton. `À Generalized Approach to Duration,'' Research in
Finance, 7 (1988): 163181.
Chambers, D. R., W. T. Carleton, and R. W. McEnally. `Ìmmunizing Default-Free Bond
Portfolios with a Duration Vector,'' Journal of Financial and Quantitative Analysis, 23, no. 1
(March 1988): 89104.
Chambers, D., W. Carleton, and D. Waldman. `À New Approach to Estimation of Terms
Structure of Interest Rates.'' Journal of Financial and Quantitative Analysis, 19, no. 3 (1984):
233252.
Chicago Board of Trade. Emergency Action, 1989 Soybeans, Chicago: Board of Trade of the
City of Chicago, 1990.
Chopra, V., and W. Ziemba. ``The Eect of Errors in Means, Variances and Covari-
ances on Optimal Portfolio Choice.'' Journal of Portfolio Management, Winter (1993):
611. Reprinted in W. T. Ziemba and J. Mulvey, Worldwide Asset and Liability Model-
ling. Cambridge, U.K.: Cambridge University Press, Publications of the Newton Institute,
1998.
Coleman, T. S., L. Fisher, and R. Ibbotson. `Èstimating the Term Structure of Interest from
Data That Include the Prices of Coupon Bonds.'' The Journal of Fixed Income, September
(1992): 85116.
Consiglio, A., and S. Zenios. `À Model for Designing Callable Bonds and Its Solution using
Tabu Research.'' Journal of Economic Dynamics and Control, 21 (1997): 14451470.
Courant, E., and D. Hilbert. Methods of Mathematical Physics, pp. 9397. New York:
Wiley, 1953.
Cox, Ingersoll, and Ross. `Àn Intertemporal General Equilibrium Model of Asset Prices.''
Econometrica, 53, no. 2 (1985): 363384.
Cox, Ingersoll, and Ross. `À Theory of the Term Structure of Interest Rates.'' Econo-
metrica, 53, no. 2 (1985): 385407.
442 References
Dolezal, J., and J. Fidler (eds.). System Modelling and Optimization, pp. 574582. New
York: Chapman and Hall, 1995.
Dorfman, R. `Àn Economic Interpretation of Optimal Control Theory.'' The American
Economic Review (1969): 817831.
Due, D. Futures Markets. Englewood Clis, N.J.: Prentice Hall, 1989, p. 99.
Dupacova, J., and M. Bertocchi. ``From Data to Model and Back to Data: A Bond Portfo-
lio Management Problem.'' In European Journal of Operations Research.
Dupacova, J., and M. Bertocchi. ``Management of Bond Portfolios via Stochastic Pro-
grammingPostoptimality and Sensitivity Analysis.'' In J. Dolezal and J. Fidler (eds.),
System Modelling and Optimization. pp. 574582. New York: Chapman and Hall, 1995.
Dym, S. ``Measuring the Risk of Foreign Bonds.'' The Journal of Portfolio Management,
Winter (1991): 5661.
Elsgolc, L. Calculus of Variations, International Series of Monographs in Pure and Applied
Mathematics. Reading, Mass.: Pergamon Press, 1962.
Elsgolts, L. Di erential Equations and the Calculus of Variations. Moscow: MIR Publishers,
1970.
Fabozzi, F. The Handbook of Treasury Securities. Probus, 1986.
Fama, E. ``Mandelbrot and the Stable Paretian Hypothesis.'' In P. Cootner, ed., The
Random Character of Stock Market Prices. Cambridge, Mass.: MIT Press, 1964.
Fama, E. ``Portfolio Analysis in a Stable Paretian Market.'' Management Science, 11 (1965).
Fama, E. ``The Behavior of Stock Market Prices.'' Journal of Business, 38 (1965).
Figlewski, S. Hedging with Financial Futures for Institutional Investors, pp. 105111.
Cambridge, Mass.: Ballinger, 1986.
Fisher, I. ``Mathematical Investigations in the Theory of Value and Price; Appreciation and
Interest'' (1896). In Reprints of Economic Classics.'' New York: A. M. Kelley, 1965.
Fisher, L., and R. Weil. ``Coping with the Risk of Interest-rate Fluctuations: Return to
Bondholders from Naive and Optimal Strategies.'' The Journal of Business, 44, no. 4
(October 1971).
Fisher, M., D. Nychka, and D. Zervos. ``Fitting the Term Structure of Interest Rates with
Smoothing Splines.'' Working Paper no. 95-1. Finance and Economics Discussion Series,
Federal Reserve Board.
Gelfand, I. M., and S. V. Fomin. Calculus of Variations. Englewood Clis, N.J.: Prentice
Hall, 1963.
Girsanov, I. V. `Òn Transforming a Certain Class of Stochastic Processes by Absolutely
Continuous Substitution of Measures.'' Theory of Probability and Its Applications, Vol. 5,
no. 3 (1960): 285301.
Golub, B., M. Holmer, R. MacKendall, and S. Zenios. ``Stochastic Programming
Models for Money Management.'' European Journal of Operational Research, 85 (1995):
282296.
Grauer, R. R., and N. H. Hakansson. ``Gains from International Diversication: 196885
Returns on Portfolios of Stocks and Bonds.'' Journal of Finance, 42 (July): 721739.
Grauer, R. R., and N. H. Hakansson. ``Higher Return, Lower Risk: Historical Returns on
Long-Run, Actively Managed Portfolios of Stocks, Bonds and Bills, 19361978.'' Financial
Analysts Journal, 38 (MarchApril 1982): 3953.
Grauer, R. R., and N. H. Hakansson. ``Returns of Levered, Actively Managed Long-Run
Portfolios of Stocks, Bonds and Bills, 19341984.'' Financial Analysts Journal, 41 (Septem-
berOctober 1985): 2443.
443 References
Grauer, R. R., and N. H. Hakansson. ``Stein and CAPM Estimators of the Means in Asset
Allocation.'' International Review of Financial Analysts, 4 (1995): 3566.
Hawawini, G. `Òn the Mathematics of Macauley's Duration.'' In G. Hawawini (ed.), Bond
Duration and Immunization: Early Developments and Recent Contributions. New York and
London: Garland Publishing, 1982.
Harrison, M., and S. Pliska. ``Martingales and Stochastic Integrals in the Theory of Con-
tinuous Trading.'' Stochastic Processes and Their Applications, 11 (1981): 215260.
Heath, D., R. Jarrow, and A. Morton. ``Bond Pricing and the Term Structure of Interest
Rates: A New Methodology for Contingent Claims Valuation.'' Econometrica, 60, no. 1
(January 1992): 77105.
Hiller, R. S., and C. Schaak. `À Classication of Structured Bond Portfolio Modeling
Techniques.'' Journal of Portfolio Management (1990): 3748.
Hines, W. G. S. ``Geometric Mean.'' In Encyclopedia of Statistical Sciences, Vol. 3, ed.
S. Kotz and N. L. Johnson, pp. 397450. New York: Wiley, 1983.
Ho, T. S. Y., and S. B. Lee. ``Term Structure Movements and Pricing Interest Rate Con-
tingent Claims.'' Journal of Finance, December (1986): 10111029.
Hull, J. Options, Futures and Derivative Securities. Englewood Clis, N.J.: Prentice Hall,
1993.
Hull, J., and A. White. ``Pricing Interest Rate Derivative Securities.'' Review of Financial
Studies, 4 (1990): 573592.
Hull, J., and A. White. `Òne-Factor Interest Rate Models and the Valuation of Interest
Rate Derivative Securities.'' Journal of Financial and Quantitative Analysis, June (1993):
235254.
Ibbotson Associates. Stocks, Bonds, Bills and Ination, Chicago: Ibbotson Associates, 1996.
Ingersoll, J. E. `Àn Examination of Convertible Securities.'' Journal of Financial Economics,
4 (1977): 289322.
Jarrow, R., and S. Turnbull. Derivative Securities. Cincinnati, Ohio: South-Western, 1996.
Jegadeesh, N. ``Treasury Auction Bids and the Salomon Squeeze.'' Journal of Finance, 48
(1993): 14031419.
Kallberg, J. G., and W. T. Ziemba. ``Mis-specication in Portfolio Selection Problems.'' In
Risk and Capital, ed. G. Bamberg and K. Spremann, Lecture Notes in Econometrics and
Mathematical Systems. New York: Springer-Verlag, 1989.
Keynes, J. M. A Treatise of Money, Vol. II, ``The Applied Theory of Money,'' Chapter 2a,
part (V). The Theory of the Forward Market. Quoted by Darrell Due in Futures Markets,
p. 99. Englewood Clis: N.J., Prentice Hall, 1989.
Khaldun, I. The Muqaddimah: An Introduction to History, New York: Bollingen, 1958
(original text in arabic, written in 1376); new English edition: Princeton: Princeton Univer-
sity Press, 1980.
Killcollin, T. E. ``Dierence Systems in Financial Futures Contracts.'' Journal of Finance
(December 1982): 11831197.
Kilman, S., and S. McMurray. ``CBOT Damaged by Last Week's Intervention.'' The Wall
Street Journal, July 13, 1987, Sec. C, pp. 1 and 12.
Kilman, S., and S. Shellenberger. ``Soybeans Fall as CBOT Order Closes Positions.'' The
Wall Street Journal, July 13, 1989, Sec. C, pp. 1 and 13.
Kolb, R. Understanding Futures Markets. Kolb Publishing Company, 1994.
Koponen, I. Physical Review, E, 52, (1995): 1197.
de La Grandville, O. ``Capital Theory, Optimal Growth and Eciency Conditions with
Exhaustible Resources.'' Econometrica, 48, no. 7 (November 1980): 17631776.
444 References
de La Grandville, O. `Ìntroducing the Geometric Moments of Continuous Random Vari-
ables, with Applications to Long-Term Growth Rates.'' Research Report, University of
Geneva, Geneva, Switzerland, 2000.
de La Grandville, O. ``New Hamiltonians for Higher-Order Equations of the Calculus of
Variations: a Generalization of the Dorfman Approach,'' Archives des Sciences, 45, no. 1
(May 1992): 5158.
de La Grandville, O. ``The Long-Term Expected Rate of Return: Setting It Right.'' Financial
Analysts Journal, November/December (1998): 7580.
de La Grandville, O. Theorie de la croissance economique. Paris: Masson, 1977.
Luenberger, D. Investment Science. Oxford: Oxford University Press, 1998.
Macaulay, F. ``Some Theoretical Problems Suggested by the Movements of Interest Rates,
Bond Yields and Stock Prices in the United States since 1865.'' National Bureau of Economic
Research.
MacBeth, J. D. ``What's the Long-Term Expected Return to your Portfolio?'' Financial
Analysts Journal, SeptemberOctober (1995): 68.
Mandelbrot, B. Fractals and Scaling in Finance: Discontinuity, Concentration, Risk. New
York: Springer, 1997.
Mandelbrot, B. ``The Variation of Certain Speculative Prices.'' Journal of Business, 36
(1963): 394419.
Maruyama, G. `Òn the Transition Probability Functions of the Markov Processes.'' Nat.
Sci. Rep. Ochanomizu Univ. 5 (1954): 1020.
Maruyama, G. ``Continuous Markov processes and stochastic equation.'' Rend. Circ. Mat.
Palermo II Ser. 4 (1955): 4890.
McCulloch, J. H. ``Measuring the Term Structure of Interest Rates.'' Journal of Business, 44
(1971): 1931.
McCulloch, J. H. ``The Tax-Adjusted Yield Curve.'' Journal of Finance, 30 (1975): 811830.
Meisner, J. F., and J. W. Labuszewski. ``Treasury Bond Futures Delivery Bias.'' Journal of
Futures Markets, Winter (1984): 569572.
Merton, R. ``Theory of Rational Option Pricing.'' Bell Journal of Economics and Manage-
ment Science, 4 (1973): 141183.
Miron, P., and P. Swannell. Pricing and Hedging Swaps. Euromoney Books, 1991.
Musiela, M., and M. Rutkowski. Martingale Methods in Financial Modelling. New York:
Springer, 1998.
Nelson, C., and S. Schaefer. ``The Dynamics of the Term Structure and Alternative Portfolio
Immunization Strategies.'' in G. O. Bierwag, G. Kaufman, and A. Toevs, eds., Innovation in
Bond Portfolio Management: Duration Analysis and Immunization. Greenwich, Conn.: JAI
Press, 1983.
Nelson, C., and A. Siegel. ``Parsimonious Modeling of Yield Curve.'' Journal of Business, 60,
no. 4 (1987): 473489.
New Palgrave Dictionary of Money and Finance. New York: Macmillan, 1992.
Peters, E. Fractal Market Analysis. New York: Wiley, 1994.
Pitman, J. Probability. New York: Springer-Verlag, 1993.
Pontryagin, L., V. Boltyanskii, R. Gamkrelidze, and E. Mishchenko. The Mathematical
Theory of Optimal Processes. New York: Interscience Publishers, 1962.
Redington, F. M. ``Review of the Principle of Life Oce Valuations.'' Journal of the Insti-
tute of Actuaries, 18 (1952): 286340. Reprinted in G. Hawawini, ed., Bond Duration and
Immunization. New York: Garland, 1982.
445 References
Reinsch, C. H. ``Smoothing by Spline Functions II,'' Numerische Mathematik, 16 (1971):
451454.
Richard, S. `Àn Arbitrage Model of the Term Structure of Interest Rates.'' Journal of
Financial Economics, 6 (1978): 3357.
Ross, S. A First Course in Probability, p. 346. New York: Macmillan, 1987.
Samuelson, P. A. `Ècient Portfolio Selection for Pareto-Levy Investments.'' Journal of
Financial and Quantitative Analysis, June (1967).
Samuelson, P. A. ``The Eect of Interest Rate Increases on the Backing System,'' The
American Economic Review, March (1945): 1627.
Schaefer, S., and E. Schwartz. `À Two-Factor Model of the Term Structure: An Approxi-
mate Solution.'' Journal of Financial and Quantitative Analysis, no. 4 (1984): 413424.
Schwarz, H. R. Numerical Analysis: A Comprehensive Introduction. New York: Wiley, 1989.
Sharpe, W. F. ``Capital Asset Prices: A Theory of Market Equilibrium under Conditions of
Risk.'' Journal of Finance, 19 (September 1964): 425442.
Sharpe, W. F., G. J. Alexander, and J. V. Bailey. Investments, 6th ed. Englewood Clis,
N.J.: Prentice Hall International, 1999.
Shea, G. ``Pitfalls in Smoothing Interest Rate Term Structure Data: Equilibrium Models and
Spline Approximations.'' Journal of Financial and Quantitative Studies, 19, no. 3 (1984):
253269.
Shiller, R. Irrational Exuberance. Princeton, N.J.: Princeton University Press, 2000.
Shiller, R. ``The Term Structure of Interest Rates.'' in B. Friedman and F. Hahn, eds.,
Handbook of Monetary Economics, North Holland, 1989.
Shorter Oxford English Dictionary, 2 Vols., p. 1540. Oxford: Oxford University Press, 1973.
Solnik, B., C. Boucrelle, and Y. Le Fur. `Ìnternational Market Correlation and Volatility.''
Financial Analysts Journal, SeptemberOctober (1996): 1734.
Solow, R. M. `À Contribution to the Theory of Economic Growth.'' The Quarterly Journal
of Economics, 70 (1956): 439469.
Sunderasan, S. M. Fixed Income Markets and Their Derivatives. Cincinnati, Ohio: South-
Western College Publishing, 1997.
Svensson, L. `Èstimating and interpreting forward interest rates: Sweden 199294.'' IMF
Working Paper, No. 114, 1994.
Svensson, L. `Èstimating forward interest rates with the extended Nelson & Siegel method.''
Sveriges Riksbank Quarterly Review, 3 (1995): 13.
Turgot de L'Aulne, A. Reexions sur la formation et la distribution des richesses. Reprinted
by Dalloz, Paris, 1957. See particularly propositions LXXXII to XCIX, pp. 128139.
Vasicek, O. `Àn Equilibrium Characterization of the Term Structure.'' Journal of Financial
Economics, 5 (November 1977): 177188.
Vasicek, O., and H. G. Fong. ``Term Structure Modeling Using Exponential Splines.''
Journal of Finance, May (1977): 339348.
Waggoner, D. ``Spline Methods for Extracting Interest Rate Curves from Coupon Bond
Prices.'' Working Paper, No. 97-10, Federal Reserve Bank of Atlanta, 1997.
Wahba, G. Spline Models for Observational Data. Philadelphia: SIAM, 1990.
Zenios, S. A. `À Model of Portfolio Management with Mortgage-backed Securities.'' Annals
of Operations Research, 43 (1993): 337356.
Ziemba, W. T., and J. Mulvey, eds. Worldwide Asset and Liability Modelling. Cambridge,
U.K.: Cambridge University Press, Publications of the Newton Institute, 1998.
446 References
INDEX
Adams, K. J. , 253
Anderson, N., 247
Anderson-Sleath model, 256257
Arak, M., 127
Arbitrage, x, xi, xii, xiv, 69
bond valuation, 2555
coupon-bearing bonds, 2934, 4952
discount factors as prices, 3436
opened- and closed-form expressions, 39
44
process, 3639
zero-coupon bonds, 2629, 3536
dened, 7
derivatives pricing, 327356
Baxter-Rennie martingale evaluation,
351356
continuous-time processes, 347351
derivative value, properties, 336338
one-step model, 328331
overvalued derivative, 333, 336
probability measure change, 344347
Q-probability measure, 338339
two-period evaluation model and
extensions, 339344
undervalued derivative, 331334
of forward and spot rates, 228229, 231
232
futures pricing, 111125
basis. See Basis
with carry costs, 117123
with interest only carry costs, 116117
potential errors, 123125
on interest rates
bond prices and, 4452, 54
forward rates, 201
Arbitrageurs, 101
ARCH processes, 406
Ask yield, 1718, 6768
Asked price versus bid price, 17
Asset prices, lognormality, 238240, 257
259, 288
Autoregressive conditional hetero-
scedasticity processes, 406
Bachelier, L., 405
Basis
convergence to zero at maturity, 112
dened, 112
properties, before maturity, 112115
sign of, 124125
Baxter, M., 338, 348, 352, 360, 378, 437, 439
Baxter-Rennie martingale evaluation, 352
356
Beltratti, A., 440
Bernoulli, J., 261
Bertocchi, M., 439
Bid price versus asked price, 17
Bierwag, G., 73
Black, F., 352, 439
Black-Scholes formula, xii, 351356
Bliss, R., 253
Bond valuation, xiii, 37, 39
arbitrage, 2555. See also Arbitrage
convexity and, 157160
duration and, 7678, 211212
at given horizon, 212213
at given term structure, 211212
immunization and. See Immunization
at variable term structure, 215218
Bonds
classication, 13, 14
convertible, 19, 2124
convexity, 153155, 162163
corporate, 1819, 20, 162
evaluation. See Bond valuation
foreign, 8790, 162
perpetual, 4041
portfolio. See Portfolios
risk ratings, 19, 20. See also Riskiness
Treasury bills, 1317
Treasury bonds, 18
Treasury notes, 1718
Bouchaud, J.-Ph., 406
Brennan, M. J., 22, 24
Brownian dierential, 363364
Brownian motions
Cameron-Martin-Girsanov theorem, 347
351
HJM model, 365
Ito 's formula, 374, 376377
Cameron, R. H., 347
Capital asset pricing model, xii
Capital gain or loss, horizon rate of return
and, 64
Capitalization rate, 194, 204
Carion, D. R., 440
Carleton, W. T., 252, 308, 439
Carrying costs
dened, 113
futures pricing, 117123
Cash account
valuation, 372373
Wiener process, 362
Cash and carry, reverse, 117, 118, 122123
Cash settlement, futures contract, 109110
CBOT. See Chicago Board of Trade
Censor, Y., 440
Central limit theorem, xiv, 408
estimating value of yearly return, 269270,
285288
Central limit theorem (cont.)
lognormality, 238239, 240, 257259, 288,
405
Chambers, D. R., 252, 308, 309
Chicago Board of Trade, conversion factor,
140
calculation, 139, 141
delivery option, 109, 126127
as a price, 127128
relative bias, 129134
Choice of the bond, 125, 126127
Chopra, V., 270
Clearing houses, 102106
Closed-form expressions
of duration, 7981
open-form expressions and, 3944
Closing out, futures position, 107111
Coleman, T. S., 252
Compound rate of return. See
Continuously compounded rate of
return; Rate of return
Consiglio, A., 440
Consol bond, 40
Continuously compounded rate of return,
222226
central limit theorem and, 285288
directional duration and, 295
application, 299, 300304
Macaulay and Fisher-Weil durations
and, 296299, 302, 303
mean and variance, estimating, 289291
n-horizon rate
uniform continuous compounding and,
291292
variance, probability distribution and,
279285
one-year return, value, variance and
probability distribution, 269279, 284
285
Continuously compounded yield, 226228
Contracts. See Forward contracts; Futures
contracts
Convenience return, 118119
Conversion factor
CBOT. See Chicago Board of Trade
delivery of grades, 108109
``true'' or ``theoretical,'' 128129
Convertible bonds, xiii, 19, 2124
Convexity, xiv, 43, 153169
bond management and, 171172
coupon-bearing bonds, 172175
portfolio development, 176180
risk measurement, 157160
zero-coupon versus coupon-bearing
bonds, 174176
calculation, 155157
dened, 155
factors, 153155, 160163
immunization and, 168169, 173174,
214215
Cornering the market, 108
Corporate bonds, 1819
convexity and, 162
Moody's ratings, 20
Coupon, 12
Coupon-bearing bonds
arbitrage-enforced valuation, 2934
formulations, 4044
interest rates and price relativity, 4952
overvalued bonds, 3234
undervalued bonds, 3032
convexity, management and, 171172
two bonds, 172175
zero-coupon bonds versus, 174176
duration and maturity, 8385
immunization, HJM model, 391399,
400401
rate of return
coupon reinvestment, 6364, 145148
horizon. See Horizon rate of return
term structure derived from, 187192
yield curve and, 187
yield to maturity, 6162, 6668
Coupon rate, 2, 7981
Coupon reinvestment, 6364, 145148
Cox, 438
Cross-hedging, 101
Daily settlements, futures markets, 102106
Debt ratings, Standard and Poor's, 19
Declare his intent, 125, 126
Default consequences, futures markets, 111
Delivery day, 126
Delivery, futures contract, 107109
T-bonds, 134139
options, 109, 125127
prot, 134139
Dempster, M. A. H., 439
Density function, 276277, 286
Derivatives
arbitrage, 327356. See also Arbitrage
directional, 295296
HJM model. See Heath-Jarrow-Morton
model
Derman, 439
Direct yield, 4950
Directional derivative, 295296
Directional duration, xiv, 295
application, 299, 300302
448 Index
nonparallel displacement, 303304
parallel displacement, 302303
Macaulay and Fisher-Weil durations and,
296299, 302, 303
Discount factors, as prices
in bond valuation, 3436
term rate structure and, 189190
Discount function, 241243, 252
Distributions
ARCH processes and, 406
fat-tailed, 288, 406
fractal, 406407
lognormal. See Lognormal distributions
probability. See Probability distributions
Doleans SDE, 376377
Dollar rate of return, 194
dened, 3536
n-year horizon rate, 279285, 288
notation, 268269
one-year return, 269279, 285287
Dorfman, R., 259, 261262
Dupacova, J., 439
Duration, xiii, 6566, 7191
bond portfolio, 8587
calculation, 7374
CBOT T-bond contract, 132134
as center of gravity, 75
conceptual importance, 7576
in bond management, 172176
in portfolio development, 176180
zero-coupon bonds, 175176
dened, 72
directional. See Directional duration
Fisher-Weil. See Fisher-Weil duration
immunization and, 7576
immunizing horizon in, 143145, 214
Macaulay's. See Macaulay's duration
modied duration, 7879
properties
bond value, 211212
coupon rate, 8081
maturity, 8385
yield to maturity, 7678, 8182
EEP agreement, 110111
Elsgolc, L., 250
End-of-the-month option, 126
Equilibrium prices, x, xi, 55, 191, 408
Equilibrium rate of return, x, 1112, 198
200
Euler equation, xi, 250251
Euler-Lagrange equation, xi, 325
Euler number, 227228, 234
Euler-Poisson equation, 250252
calculus of variations and, 325
derivation, xi, 259264
Exchange for physicals, 110111
Fabozi, F., 437
Face value, 1
Fat-tailed distributions, 288, 406
Financial institutions, investment
protection, 163169
Fisher, I., xi, 46
Fisher, L., 73, 252, 254
Fisher equation, xi, xii, 4445, 4647
Fisher-Nychka-Zervos model, 253256, 257
Fisher-Weil duration, xiv, 209, 218
classical versus HJM portfolios, 395, 399
directional duration and, 296299, 302,
303
Fixed exchange rates, ix
Fixed rate of return, ix, 12, 13
Fluctuation
rate of return
bond pricing and, 52, 53, 54
nancial institutions and, 163169
yield, 23
Fong, G., 252
Foreign bonds
convexity, 162
riskiness, 8790
Forward contracts
arbitrage, 112113
characteristics and deciencies, 9699
closing out, 110
dened, 4
synthetic, 910
Forward rate curve, splines and
Fisher-Nychka-Zervos model, 253
Waggoner model, 253256
Forward rate of return, xiv
arbitrage and, xiii, 69
derived from term structure, 192198
risk averse approach, 201207
risk neutral approach, 195201, 206
model
in search for term structure, 240243
Nelson-Siegel model, 244247
parametric methods, 244248
spline methods, 248257
Svensson model, 247248
449 Index
Forward rate of return (cont.)
spot rates and, 238240
in instantaneous lending and borrowing,
228232
lognormality, 238240
in search for term structure, 240257
zero-coupon bond price and, 232233
Wiener process, 361
Fractal theory, 406407
Fractional numbers, 56
Friedman, B., 437
Functionals, 308
Functions
density, 276277, 286
discount, 241243, 252
fractal distributions, 406
functionals and, 308
Laguerre function, 244, 245
pedagogy and, 407408
Future value. See also Horizon rate of
return; Long horizon rate of return
horizon rate of return, 65
one-year return, estimating, 269279, 284
285
zero-coupon bond, forward and spot rates
and, 232233
Futures contracts
arbitrage, 111125
basis. See Basis
with carry costs, 117123
with interest only carry costs, 116117
potential errors, 123125
closing out, 107
cash settlement, 109110
default consequences, 111
delivery, 107109
exchange for physicals, 110111
reversing trades, 110
HJM applications, 401
immunization with, 168169
riskiness, 106107
standardization, 102
T-bond. See Treasury bonds
Futures markets
contracts. See Futures contracts
daily settlements, 102106
nancial instruments, 100
margins, 102106
operators, 100101
safeguards, 102
trading categories, 100
Geometric processes. See Wiener process
Girsanov, I. V., 348
Gold reserves, ix
Golub, B., 440
Goodman, L. S., 127
Grades, delivery, 107109
Grandville, O. de La, 287
Grauer, R. R., 271
Gross return, 3536
Growth theory, variational problems in, xi
Hahn, F., 437
Hakansson, N. H., 271
Hamiltonians, Euler-Poisson equation and,
259264
Harrison, M., xii
Hawawini, G., 85
Heath, D., xii, 360, 439
Heath-Jarrow-Morton model, x, xiv, 359
402
cash account, valuation, 372373
derivatives pricing, 401
immunization
Ito 's formula, with applications, 374378
one-factor model, 360361
cash account, 362, 372373
current short rate, 361362
forward rate, 361
no-arbitrage condition, 365367
Vasicek volatility reduction factor, 368
372
zero-coupon bond, 388
Hedgers, 100101
Hedging. See also Immunization
coupon-bearing bonds, 391399, 400, 401
cross-hedging, 101
forward contract, 9699
Hicks, J., 73
Hiller, R. S., 440
model
Ho, T. S. Y., 438
Holmer, R., 440
Holding period rate of return. See Horizon
rate of return
Horizon, 4, 408
Horizon rate of return, xiii
bond valuation at, 212213
coupon reinvestment, 6364, 145148
forward rates, derivation
borrowers and, 203207
investors and, 201203
450 Index
fractional numbers, 56
future value, 59, 6266
central limit theorem and, 287288
n-year, estimating, 279285
one year or less, 45
immunization. See Immunization
integer numbers, 5
long return. See Long horizon rate of
return
short return, 145, 146
borrowers and, 203
one year or less, 45
spot rate and, 185187
Hull, J., 438, 439
Ibbotson, R., 252
Immunization, xiiixiv, 143150, 209218,
383401
bond value and
at given term structure, 211212
at horizon H, 212213
at variable term structure, 215218
conditions, 213215
convexity and, 168169, 214215
management strategy, 173174
dened, 144
duration and, 7576
coupon-bearing bonds, 395, 396399,
400401
immunizing horizon in, 143145, 214
HJM applications. See Heath-Jarrow-
Morton model
likelihood, 145148
term structure variation and, 216
defensive management, 173176
diculties with, 215218
general theorem, with applications, 308
324
Indices, cash settlement and, 109110
Ination, interest rates and, x, 1011
Ingersoll, J. E., 22, 438
Initial margin, 104
Integer numbers, horizons and, 5
Intention day, notice of, 125
Interest rate. See Rate of return
Interest rate futures, 100
Internal rate of return
ask yield and, 18
Macaulay's duration, 7374
spot rate and, 186
yield to maturity and, 60
International bonds
convexity, 162
riskiness, 8790
IRR. See Internal rate of return
Ito 's formula
applications, 376378
derivation, 374375
Ito 's lemma, 363364, 375, 377378
Jarrow, R., xii, 360, 372, 387, 437, 439
Kallberg, J., 271
Kaufman, G., 73
Keynes, J. M., 55, 123125
Khaldun, I., 408
Killcollin, T., 127
Kolb, R., 126
Koponen, I., 407
Labuszewski, J., 127
Laguerre function, 244, 245
Last trading day, 126
Laurent, A., 440
Lee, S. B., 438
L'Ho pital's rule, 245, 259, 369
Lintner, J., xii
Lognormal distribution, xiv, 238240
central limit theorem and, 257259, 285
288
general use, 405
one-year return, 272277
Long forward contract, 98, 99
Long horizon rate of return, xiv, 267292
288
forward rates, derivation
borrowers and, 203205
investors, 202, 203
immunization, 145150
lognormality and, 285289
mean and variance, estimating, 289
291
n-year horizon return
probability distribution, 283285
variance, 279283
probability distribution, 272279
variance, 267272
uniform continuous compounding and,
291292
Luenberger, D., 437
Macaulay, F., 73, 163
Macaulay's duration, xiv, 7374
directional duration and, 296299
451 Index
Macaulay's duration (cont.)
immunization and, 209, 218
riskiness measurement and, 166
MacKendall, R., 440
MacLaurin approximation, 68
MacLaurin series, 309
Maintenance margin, 103104
Management style
active, 171, 174, 176
defensive, 171, 176
Mandelbrot, B., 288, 405
Margins, 102106
Markovitz, H., xii
Martin, W. T., 348
Martingale probability, xii, 329
Baxter-Rennie evaluation, 351356
HJM model, 363365, 367368, 401
Ito 's lemma and, 377378
P-martingale, 345
Q-probability and, 338339, 347
Maturity
in bond valuation, 4144, 54
convergence of basis to zero at, 112
dened, 1
duration and, 8285
prot from delivery and, 138
properties of basis before, 112115
yield to maturity, 5962
McCulloch, J. H., 252
McEnally, R. W., 308, 439
Meisner, J., 127
Merton, R., xii
Modied duration, 7879
Moriggia, V., 439
Moody's corporate bond ratings, 20
Morton, A., xii, 360, 439
Mulvey, J., 439
Musiela, M., 437
Nelson, C., 244, 438
Nelson-Siegel model, 244245, 246, 247
Notice of intention day, 125
Numbers
Euler number, 227228, 234
fractional, 56
integer, 5
Nychka, D., 253
Osetting positions, 110
Oil shocks of 1973 and 1979, x
One-year return, estimating value, variance
and probability distribution, 269279,
284285
Open-form expressions
closed-form expressions and, 3944
of duration, 8182
Opportunity cost, 113
Ostrogradski equation, 264, 325
P-martingale, 345
P-measure, 344, 345, 347, 352
P-probability measure, 344345
Par value, 1
Parametric methodology, 246, 247
versus spline methodology, 256257
Perpetual bond, 4041
Peters, E., 406, 407
Physicals, exchange for, 110111
Pliska, S. R., xii, 439
Polynomials. See Spline methodology
Pontryagin's principle, 260261
Portfolios
convexity, 153155, 160163
in portfolio development, 176180
duration, 8587, 143144
model
immunization, 143144, 160163
coupon-bearing bonds, 395, 396399,
400401
variable term structure and, 215218
391
Position day, 125
Potters, M., 406
Price
arbitrage. See Arbitrage
bid versus asked, 17
Capital asset pricing model, xii
CBOT conversion factor as, 127128
determination, 5455
discount factors as
term rate structure and, 189190
in valuation, 3436
equilibrium, 55, 191, 408
uctuating rate of return and, 52, 53, 54
model
lognormality, 238240, 257259, 288
sensitivity, duration and, 7678
Principal, 1
Probability distribution
288
independence, 406
452 Index
n-year horizon return, 283285
Probability measures, xiv
change in, 344347, 363365
Martingale, 329, 338339
Q-probability measure. See Q-probability
measure
Prot from delivery, 134139
Pure-discount security, 2
Q-martingale process, 353355
Q-measure
Black-Scholes formula and, 352356
martingale, 345, 347
Q-probability measure, 338339
Black-Scholes formula and, 352356
351
in two-period evaluation model and
extensions, 342345
Quality option, 125, 126127
Radon-Nikodym process, 345, 346, 347
Rate of return, ix, xiii, 324. See also Yield
arbitrage, 69
bond prices and, 4452, 53, 54
forward rates, 201
in bond valuation, 4144
compound. See Continuously
compounded rate of return
convexity. See Convexity
dollar rate, 194
dened, 3536
n-year horizon rate, 279285, 288
notation, 268269
one-year return, 269279, 285287
equilibrium rate, 1112, 198200
xed rate, 12, 13
forward rate. See Forward rate of return
horizon rate. See Horizon rate of return;
Long horizon rate of return
immunization. See Immunization
internal rate
ask yield and, 18
Macaulay's duration, 7374
spot rate and, 186
yield to maturity and, 60
long-term. See Long horizon rate of return
rationales for, 1013
spot rate, 4, 185187
term structure. See Term structure
volatility of, xiii
Redington, F. M., 73, 76, 163
Redington conditions, 164168
Redington property, 164
Reinvestment, of coupons, 6364, 145148
Reinvestment risk, 3
Rennie, A., 338, 348, 352, 360, 378, 437
Reverse cash and carry, 117, 122123
Reversing trades, 110
Richard, S., 438
Risk premium, 4748, 204
Riskiness
foreign bonds, 8790
forward contracts, 9699
futures trading, 106107
long horizon forward rates
borrowers, 203207
investors, 202203
measures, xii
duration, 7779
Moody's corporate ratings, 20
S&P debt ratings, 19
risk premium, xii, 4748, 204
short horizon forward rates
borrowers, 203
investors, 201202
Ross, S., 127, 257, 438
Roughness penalty, 253256
Rutkowski, M., 437
Samuelson, P., 73
Savings and loan association failures, 165
Scalpers, 101
Schaak, C., 440
Schaefer, S., 438
Schwartz, E. S., 22, 24, 438
SDE, Doleans, 376377
Securities, pure-discount, 2
Settlement price, 125126
Sharpe, W., xii, 48
Shea, G., 252
Shiller, R., 406, 437
Short (the), 97
Short forward contract, 9798
Short horizon rate of return, 145, 146
borrowers and, 203
Wiener process, 361362
Siegel, A., 244
Simple interest rate, continuous
compounding and, 222225
Sleath, J., 247
Solow, R., xi, 407
S&P. See Standard and Poor's
453 Index
Speculators, 101
Spline, 248249
Spline methodology, 248253
versus parametric methodology, 256
257
Spot price
forward contracts, 9798
futures trading
arbitrage, 101, 111125. See also
Arbitrage
speculators, 101
Spot rate curve. See Term structure
Spot rate of return
arbitrage, xiv, 69
dened, 4, 185187
forward rates and. See Forward rate of
return
immunization and. See Immunization
structure. See Term structure
Standard and Poor's, 19, 109110
Stocks
arbitrage pricing. See Arbitrage,
derivatives pricing
convertible bonds and, 19, 2122
Sundaresan, S., 437
Taylor approximation, 68
Taylor development, 224
Taylor expansion, 158, 312
Taylor series
convexity calculations, 157
immunization calculations, 309, 312, 325
shock sensitivity, 384
Term structure, xiixiii, xiv
in bond valuation, 37, 55
given, 211212
at horizon H, 212213
variable, 215218
calculation, 187
dened, 187
forward rates derived from, 192198
risk averse approach, 201207
risk neutral approach, 195201, 206
parallel changes in, xiv
of rates, search for, 240243
parametric methods, 244248
spline methods, 248257
variation, immunization and. See also
Directional duration
defensive management, 173174, 175,
176
diculties with, 215218
general theorem, with applications, 308
324
Theorems
Cameron-Martin-Girsanov, 347351
central limit. See Central limit theorem
immunization, with applications, 308
324
Time to deliver option, 125126
Toevs, A., 73
Toy, 439
Trade balance, ix
Trading Act of 1921, 108
Trading day, last, 126
Trading period, 4, 408
Treasury bills, 1317, 401
Treasury bonds, 14, 18
Treasury bonds futures contract, 125139
conversion factor
CBOT. See Chicago Board of Trade,
conversion factor
true or theoretical, 128129
delivery options, 109, 125127
delivery prot, 134139
Treasury notes, 14, 1718
Turgot de L'Aulne, A., x, xi, xii
Turgot-Fisher equation, xi, 47
Turnbull, S., 371, 387, 437
Valuation. See Bond valuation
Van Deventer, D. R., 253
Variable roughness penalty, 253256
Variance
mean and, estimating, 289291
n-year horizon return, 279283
Vasicek, O., 252, 368, 438
372
Volatility reduction factor
classical versus HJM portfolios, 389391,
395, 401
Vasicek's, 368372
Volatility term, xiii
VRP, 253256
454 Index
Waggoner, D., 248, 253
Waldman, D., 252
Weil, R., 73
White, A., 438, 439
Wiener process, xiii, 405
cash account as, 362
forward rate as, 361
lognormality, 239240
short rate as, 361362
zero-coupon bond as, 363
Wild card option, 125127
Yield
continuously compounded, 226228
direct, 4950
uctuation, 23
Yield curve, 183, 187192
Yield to maturity, xiii, 5962
ask yield, 1718, 6768
dened, 183184
as performance measure, 62
semi-annual coupons, 6162, 6668
Zenios, S., 439, 440
Zero-coupon bonds
dened, 2
duration and maturity, 83
forward and spot rates and, 232233
HJM model. See also Heath-Jarrow-
Morton model
immunization, 387389, 390391
one-factor model, 363
protection, 384387
management, convexity in, 174176
term structure derived from, 188, 191, 191.
See also Term structure
valuation, 2629, 3536, 3839
overvalued bonds, 2829
undervalued bonds, 2728
Wiener processes, 363, 364
Zervos, D., 253
Ziemba, W., 270, 271, 439
455 Index

Olivier de La Grandville-Bond Pricing and Portfolio Analysis-The MIT Press (2000)

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Olivier de La Grandville-Bond Pricing and Portfolio Analysis-The MIT Press (2000)

Diunggah oleh

Hak Cipta:

Format Tersedia

M

, in the same way that closed

The derivative with respect to i is

line (equal to S(T)). He incurs considerable risk

line going through p(tY T), equal to S(T) p(tY T).

Yield of bond with maturity

, will therefore be the median of the log-

= 1X0822) and the mean of X

where u is replaced, in the right-hand side, by u = log x and dxadu =

. In our case, however, since x = e

, and it is a direct consequence of Jensen's

, as well as the intervals corresponding to the intervals I

. Repeat this process for a new interval of the same length,

, it is clear that the probability of having an X value

than it was when

oZY Z@ N(0Y 1) (13)

1); in turn, and if needed, this interval can

1) = (3X2%Y 21X0%). On the

R(t) = 0X048 0X00146 (5X5)10

line in quadrant III)

Ih. Our purpose is to show that

plays the role of the parameter. We have

355 Arbitrage Pricing in Discrete and Continuous Time

oZ, Z@ N(0Y 1) and from

Anda mungkin juga menyukai